Merge branch 'main' into benchmark-format-args

.github/workflows/ci.yaml

@@ -156,7 +156,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        os: [ubuntu-latest, macos-14]
+        os: [ubuntu-latest, macos-14, windows-latest]
         python-version: ["3.11"]
         torch: [latest]
         include:

.github/workflows/docs.yml

@@ -46,7 +46,7 @@ jobs:
         run: pip install ruff black tqdm mkdocs-material "mkdocstrings[python]" mkdocs-jupyter mkdocs-redirects mkdocs-ultralytics-plugin mkdocs-macros-plugin
       - name: Ruff fixes
         continue-on-error: true
-        run: ruff check --fix --fix-unsafe --select D --ignore=D100,D104,D203,D205,D212,D213,D401,D406,D407,D413 .
+        run: ruff check --fix --unsafe-fixes --select D --ignore=D100,D104,D203,D205,D212,D213,D401,D406,D407,D413 .
       - name: Update Docs Reference Section and Push Changes
         if: github.event_name == 'pull_request_target'
         run: |

.github/workflows/publish.yml

@@ -34,7 +34,7 @@ jobs:
       - name: Install dependencies
         run: |
           python -m pip install --upgrade pip wheel
-          pip install openai requests build twine toml
+          pip install requests build twine toml
       - name: Check PyPI version
         shell: python
         run: |
@@ -79,8 +79,8 @@ jobs:
               publish = True  # First release
 
           os.system(f'echo "increment={publish}" >> $GITHUB_OUTPUT')
-          os.system(f'echo "version={local_version}" >> $GITHUB_OUTPUT')
-          os.system(f'echo "previous_version={online_version or "N/A"}" >> $GITHUB_OUTPUT')
+          os.system(f'echo "current_tag=v{local_version}" >> $GITHUB_OUTPUT')
+          os.system(f'echo "previous_tag=v{online_version}" >> $GITHUB_OUTPUT')
 
           if publish:
               print('Ready to publish new version to PyPI ✅.')
@@ -88,81 +88,18 @@ jobs:
       - name: Publish new tag
         if: (github.event_name == 'push' || github.event.inputs.pypi == 'true')  && steps.check_pypi.outputs.increment == 'True'
         run: |
-          git tag -a "v${{ steps.check_pypi.outputs.version }}" -m "$(git log -1 --pretty=%B)"  # i.e. "v0.1.2 commit message"
-          git push origin "v${{ steps.check_pypi.outputs.version }}"
+          git tag -a "${{ steps.check_pypi.outputs.current_tag }}" -m "$(git log -1 --pretty=%B)"  # i.e. "v0.1.2 commit message"
+          git push origin "${{ steps.check_pypi.outputs.current_tag }}"
       - name: Publish new release
         if: (github.event_name == 'push' || github.event.inputs.pypi == 'true')  && steps.check_pypi.outputs.increment == 'True'
         env:
           OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
           GITHUB_TOKEN: ${{ secrets.PERSONAL_ACCESS_TOKEN || secrets.GITHUB_TOKEN }}
-          CURRENT_TAG: ${{ steps.check_pypi.outputs.version }}
-          PREVIOUS_TAG: ${{ steps.check_pypi.outputs.previous_version }}
-        shell: python
+          CURRENT_TAG: ${{ steps.check_pypi.outputs.current_tag }}
+          PREVIOUS_TAG: ${{ steps.check_pypi.outputs.previous_tag }}
         run: |
-          import openai
-          import os
-          import requests
-          import json
-          import subprocess
-
-          # Retrieve environment variables
-          OPENAI_API_KEY = os.getenv('OPENAI_API_KEY')
-          GITHUB_TOKEN = os.getenv('GITHUB_TOKEN')
-          CURRENT_TAG = os.getenv('CURRENT_TAG')
-          PREVIOUS_TAG = os.getenv('PREVIOUS_TAG')
-
-          # Check for required environment variables
-          if not all([OPENAI_API_KEY, GITHUB_TOKEN, CURRENT_TAG, PREVIOUS_TAG]):
-              raise ValueError("One or more required environment variables are missing.")
-
-          latest_tag = f"v{CURRENT_TAG}"
-          previous_tag = f"v{PREVIOUS_TAG}"
-          repo = os.getenv('GITHUB_REPOSITORY')
-          headers = {"Authorization": f"token {GITHUB_TOKEN}", "Accept": "application/vnd.github.v3.diff"}
-
-          # Get the diff between the tags
-          url = f"https://api.github.com/repos/{repo}/compare/{previous_tag}...{latest_tag}"
-          response = requests.get(url, headers=headers)
-          diff = response.text if response.status_code == 200 else f"Failed to get diff: {response.content}"
-
-          # Get summary
-          messages = [
-              {
-                  "role": "system",
-                  "content": "You are an Ultralytics AI assistant skilled in software development and technical communication. Your task is to summarize GitHub releases in a way that is detailed, accurate, and understandable to both expert developers and non-expert users. Focus on highlighting the key changes and their impact in simple and intuitive terms."
-              },
-              {
-                  "role": "user",
-                  "content": f"Summarize the updates made in the '{latest_tag}' tag, focusing on major changes, their purpose, and potential impact. Keep the summary clear and suitable for a broad audience. Add emojis to enliven the summary. Reply directly with a summary along these example guidelines, though feel free to adjust as appropriate:\n\n"
-                             f"## 🌟 Summary (single-line synopsis)\n"
-                             f"## 📊 Key Changes (bullet points highlighting any major changes)\n"
-                             f"## 🎯 Purpose & Impact (bullet points explaining any benefits and potential impact to users)\n"
-                             f"\n\nHere's the release diff:\n\n{diff[:300000]}",
-              }
-          ]
-          client = openai.OpenAI(api_key=OPENAI_API_KEY)
-          completion = client.chat.completions.create(model="gpt-4o-2024-08-06", messages=messages)
-          summary = completion.choices[0].message.content.strip()
-
-          # Get the latest commit message
-          commit_message = subprocess.run(['git', 'log', '-1', '--pretty=%B'], check=True, text=True, capture_output=True).stdout.split("\n")[0].strip()
-
-          # Prepare release data
-          release = {
-              'tag_name': latest_tag,
-              'name': f"{latest_tag} - {commit_message}",
-              'body': summary,
-              'draft': False,
-              'prerelease': False
-          }
-
-          # Create the release on GitHub
-          release_url = f"https://api.github.com/repos/{repo}/releases"
-          release_response = requests.post(release_url, headers=headers, data=json.dumps(release))
-          if release_response.status_code == 201:
-              print(f'Successfully created release {latest_tag}')
-          else:
-              print(f'Failed to create release {latest_tag}: {release_response.content}')
+          curl -s "https://raw.githubusercontent.com/ultralytics/actions/main/utils/summarize_release.py" | python -
+        shell: bash
       - name: Publish to PyPI
         continue-on-error: true
         if: (github.event_name == 'push' || github.event.inputs.pypi == 'true')  && steps.check_pypi.outputs.increment == 'True'
@@ -193,7 +130,7 @@ jobs:
         uses: slackapi/slack-github-action@v1.26.0
         with:
           payload: |
-            {"text": "<!channel> GitHub Actions success for ${{ github.workflow }} ✅\n\n\n*Repository:* https://github.com/${{ github.repository }}\n*Action:* https://github.com/${{ github.repository }}/actions/runs/${{ github.run_id }}\n*Author:* ${{ github.actor }}\n*Event:* NEW '${{ github.repository }} v${{ steps.check_pypi.outputs.version }}' pip package published 😃\n*Job Status:* ${{ job.status }}\n*Pull Request:* <https://github.com/${{ github.repository }}/pull/${{ env.PR_NUMBER }}> ${{ env.PR_TITLE }}\n"}
+            {"text": "<!channel> GitHub Actions success for ${{ github.workflow }} ✅\n\n\n*Repository:* https://github.com/${{ github.repository }}\n*Action:* https://github.com/${{ github.repository }}/actions/runs/${{ github.run_id }}\n*Author:* ${{ github.actor }}\n*Event:* NEW '${{ github.repository }} ${{ steps.check_pypi.outputs.current_tag }}' pip package published 😃\n*Job Status:* ${{ job.status }}\n*Pull Request:* <https://github.com/${{ github.repository }}/pull/${{ env.PR_NUMBER }}> ${{ env.PR_TITLE }}\n"}
         env:
           SLACK_WEBHOOK_URL: ${{ secrets.SLACK_WEBHOOK_URL_YOLO }}
       - name: Notify on Slack (Failure)

docs/build_docs.py

@@ -164,7 +164,7 @@ def update_docs_html():
     # Convert plaintext links to HTML hyperlinks
     files_modified = 0
     for html_file in tqdm(SITE.rglob("*.html"), desc="Converting plaintext links"):
-        with open(html_file, "r", encoding="utf-8") as file:
+        with open(html_file, encoding="utf-8") as file:
             content = file.read()
         updated_content = convert_plaintext_links_to_html(content)
         if updated_content != content:

docs/en/datasets/detect/index.md

@@ -77,6 +77,7 @@ Here is a list of the supported datasets and a brief description for each:
 - [COCO](coco.md): Common Objects in Context (COCO) is a large-scale object detection, segmentation, and captioning dataset with 80 object categories.
 - [LVIS](lvis.md): A large-scale object detection, segmentation, and captioning dataset with 1203 object categories.
 - [COCO8](coco8.md): A smaller subset of the first 4 images from COCO train and COCO val, suitable for quick tests.
+- [COCO128](coco.md): A smaller subset of the first 128 images from COCO train and COCO val, suitable for tests.
 - [Global Wheat 2020](globalwheat2020.md): A dataset containing images of wheat heads for the Global Wheat Challenge 2020.
 - [Objects365](objects365.md): A high-quality, large-scale dataset for object detection with 365 object categories and over 600K annotated images.
 - [OpenImagesV7](open-images-v7.md): A comprehensive dataset by Google with 1.7M train images and 42k validation images.

docs/en/datasets/index.md

@@ -38,6 +38,7 @@ Bounding box object detection is a computer vision technique that involves detec
 - [COCO](detect/coco.md): Common Objects in Context (COCO) is a large-scale object detection, segmentation, and captioning dataset with 80 object categories.
 - [LVIS](detect/lvis.md): A large-scale object detection, segmentation, and captioning dataset with 1203 object categories.
 - [COCO8](detect/coco8.md): A smaller subset of the first 4 images from COCO train and COCO val, suitable for quick tests.
+- [COCO128](detect/coco.md): A smaller subset of the first 128 images from COCO train and COCO val, suitable for tests.
 - [Global Wheat 2020](detect/globalwheat2020.md): A dataset containing images of wheat heads for the Global Wheat Challenge 2020.
 - [Objects365](detect/objects365.md): A high-quality, large-scale dataset for object detection with 365 object categories and over 600K annotated images.
 - [OpenImagesV7](detect/open-images-v7.md): A comprehensive dataset by Google with 1.7M train images and 42k validation images.
@@ -56,6 +57,7 @@ Instance segmentation is a computer vision technique that involves identifying a
 
 - [COCO](segment/coco.md): A large-scale dataset designed for object detection, segmentation, and captioning tasks with over 200K labeled images.
 - [COCO8-seg](segment/coco8-seg.md): A smaller dataset for instance segmentation tasks, containing a subset of 8 COCO images with segmentation annotations.
+- [COCO128-seg](segment/coco.md): A smaller dataset for instance segmentation tasks, containing a subset of 128 COCO images with segmentation annotations.
 - [Crack-seg](segment/crack-seg.md): Specifically crafted dataset for detecting cracks on roads and walls, applicable for both object detection and segmentation tasks.
 - [Package-seg](segment/package-seg.md): Tailored dataset for identifying packages in warehouses or industrial settings, suitable for both object detection and segmentation applications.
 - [Carparts-seg](segment/carparts-seg.md): Purpose-built dataset for identifying vehicle parts, catering to design, manufacturing, and research needs. It serves for both object detection and segmentation tasks.
@@ -88,6 +90,7 @@ Image classification is a computer vision task that involves categorizing an ima
 Oriented Bounding Boxes (OBB) is a method in computer vision for detecting angled objects in images using rotated bounding boxes, often applied to aerial and satellite imagery.
 
 - [DOTA-v2](obb/dota-v2.md): A popular OBB aerial imagery dataset with 1.7 million instances and 11,268 images.
+- [DOTA8](obb/dota8.md): A smaller subset of the first 8 images from the DOTAv1 split set, 4 for training and 4 for validation, suitable for quick tests.
 
 ## [Multi-Object Tracking](track/index.md)
 

docs/en/datasets/segment/index.md

@@ -93,6 +93,7 @@ The `train` and `val` fields specify the paths to the directories containing the
 
 - [COCO](coco.md): A comprehensive dataset for object detection, segmentation, and captioning, featuring over 200K labeled images across a wide range of categories.
 - [COCO8-seg](coco8-seg.md): A compact, 8-image subset of COCO designed for quick testing of segmentation model training, ideal for CI checks and workflow validation in the `ultralytics` repository.
+- [COCO128-seg](coco.md): A smaller dataset for instance segmentation tasks, containing a subset of 128 COCO images with segmentation annotations.
 - [Carparts-seg](carparts-seg.md): A specialized dataset focused on the segmentation of car parts, ideal for automotive applications. It includes a variety of vehicles with detailed annotations of individual car components.
 - [Crack-seg](crack-seg.md): A dataset tailored for the segmentation of cracks in various surfaces. Essential for infrastructure maintenance and quality control, it provides detailed imagery for training models to identify structural weaknesses.
 - [Package-seg](package-seg.md): A dataset dedicated to the segmentation of different types of packaging materials and shapes. It's particularly useful for logistics and warehouse automation, aiding in the development of systems for package handling and sorting.

docs/mkdocs_github_authors.yaml

@@ -1,46 +1,120 @@
-116908874+jk4e@users.noreply.github.com: jk4e
-1185102784@qq.com: Laughing-q
-130829914+IvorZhu331@users.noreply.github.com: IvorZhu331
-135830346+UltralyticsAssistant@users.noreply.github.com: UltralyticsAssistant
-1579093407@qq.com: YOLOv5-Magic
-17216799+ouphi@users.noreply.github.com: ouphi
-17316848+maianumerosky@users.noreply.github.com: maianumerosky
-34196005+fcakyon@users.noreply.github.com: fcakyon
-37276661+capjamesg@users.noreply.github.com: capjamesg
-39910262+ChaoningZhang@users.noreply.github.com: ChaoningZhang
-40165666+berry-ding@users.noreply.github.com: berry-ding
-47978446+sergiuwaxmann@users.noreply.github.com: sergiuwaxmann
-48149018+zhixuwei@users.noreply.github.com: zhixuwei
-49699333+dependabot[bot]@users.noreply.github.com: dependabot
-52826299+Chayanonjackal@users.noreply.github.com: Chayanonjackal
-53246858+hasanghaffari93@users.noreply.github.com: hasanghaffari93
-60036186+mfloto@users.noreply.github.com: mfloto
-61612323+Laughing-q@users.noreply.github.com: Laughing-q
-62214284+Burhan-Q@users.noreply.github.com: Burhan-Q
-68285002+Kayzwer@users.noreply.github.com: Kayzwer
-75611662+tensorturtle@users.noreply.github.com: tensorturtle
-78843978+Skillnoob@users.noreply.github.com: Skillnoob
-79740115+0xSynapse@users.noreply.github.com: 0xSynapse
-Francesco.mttl@gmail.com: ambitious-octopus
-abirami.vina@gmail.com: abirami-vina
-ahmelsamahy@gmail.com: Ahelsamahy
-andrei.kochin@intel.com: andrei-kochin
-ayush.chaurarsia@gmail.com: AyushExel
-chr043416@gmail.com: RizwanMunawar
-glenn.jocher@ultralytics.com: glenn-jocher
-hnliu_2@stu.xidian.edu.cn: null
-jpedrofonseca_94@hotmail.com: null
-k-2feng@hotmail.com: null
-lakshantha@ultralytics.com: lakshanthad
-lakshanthad@yahoo.com: lakshanthad
-muhammadrizwanmunawar123@gmail.com: RizwanMunawar
-not.committed.yet: null
-plashchynski@gmail.com: plashchynski
-priytosh.revolution@live.com: priytosh-tripathi
-rulosanti@gmail.com: null
-shuizhuyuanluo@126.com: null
-sometimesocrazy@gmail.com: null
-stormsson@users.noreply.github.com: stormsson
-waxmann.sergiu@me.com: sergiuwaxmann
-web@ultralytics.com: UltralyticsAssistant
-xinwang614@gmail.com: GreatV
+116908874+jk4e@users.noreply.github.com:
+  avatar: https://avatars.githubusercontent.com/u/116908874?v=4
+  username: jk4e
+130829914+IvorZhu331@users.noreply.github.com:
+  avatar: https://avatars.githubusercontent.com/u/130829914?v=4
+  username: IvorZhu331
+135830346+UltralyticsAssistant@users.noreply.github.com:
+  avatar: https://avatars.githubusercontent.com/u/135830346?v=4
+  username: UltralyticsAssistant
+1579093407@qq.com:
+  avatar: https://avatars.githubusercontent.com/u/160490334?v=4
+  username: YOLOv5-Magic
+17216799+ouphi@users.noreply.github.com:
+  avatar: https://avatars.githubusercontent.com/u/17216799?v=4
+  username: ouphi
+17316848+maianumerosky@users.noreply.github.com:
+  avatar: https://avatars.githubusercontent.com/u/17316848?v=4
+  username: maianumerosky
+34196005+fcakyon@users.noreply.github.com:
+  avatar: https://avatars.githubusercontent.com/u/34196005?v=4
+  username: fcakyon
+37276661+capjamesg@users.noreply.github.com:
+  avatar: https://avatars.githubusercontent.com/u/37276661?v=4
+  username: capjamesg
+39910262+ChaoningZhang@users.noreply.github.com:
+  avatar: https://avatars.githubusercontent.com/u/39910262?v=4
+  username: ChaoningZhang
+40165666+berry-ding@users.noreply.github.com:
+  avatar: https://avatars.githubusercontent.com/u/40165666?v=4
+  username: berry-ding
+47978446+sergiuwaxmann@users.noreply.github.com:
+  avatar: https://avatars.githubusercontent.com/u/47978446?v=4
+  username: sergiuwaxmann
+48149018+zhixuwei@users.noreply.github.com:
+  avatar: https://avatars.githubusercontent.com/u/48149018?v=4
+  username: zhixuwei
+49699333+dependabot[bot]@users.noreply.github.com:
+  avatar: https://avatars.githubusercontent.com/u/27347476?v=4
+  username: dependabot[bot]
+53246858+hasanghaffari93@users.noreply.github.com:
+  avatar: https://avatars.githubusercontent.com/u/53246858?v=4
+  username: hasanghaffari93
+60036186+mfloto@users.noreply.github.com:
+  avatar: https://avatars.githubusercontent.com/u/60036186?v=4
+  username: mfloto
+61612323+Laughing-q@users.noreply.github.com:
+  avatar: https://avatars.githubusercontent.com/u/61612323?v=4
+  username: Laughing-q
+62214284+Burhan-Q@users.noreply.github.com:
+  avatar: https://avatars.githubusercontent.com/u/62214284?v=4
+  username: Burhan-Q
+68285002+Kayzwer@users.noreply.github.com:
+  avatar: https://avatars.githubusercontent.com/u/68285002?v=4
+  username: Kayzwer
+75611662+tensorturtle@users.noreply.github.com:
+  avatar: https://avatars.githubusercontent.com/u/75611662?v=4
+  username: tensorturtle
+78843978+Skillnoob@users.noreply.github.com:
+  avatar: https://avatars.githubusercontent.com/u/78843978?v=4
+  username: Skillnoob
+79740115+0xSynapse@users.noreply.github.com:
+  avatar: https://avatars.githubusercontent.com/u/79740115?v=4
+  username: 0xSynapse
+Francesco.mttl@gmail.com:
+  avatar: https://avatars.githubusercontent.com/u/3855193?v=4
+  username: ambitious-octopus
+abirami.vina@gmail.com:
+  avatar: https://avatars.githubusercontent.com/u/25847604?v=4
+  username: abirami-vina
+ahmelsamahy@gmail.com:
+  avatar: https://avatars.githubusercontent.com/u/10195309?v=4
+  username: Ahelsamahy
+andrei.kochin@intel.com:
+  avatar: https://avatars.githubusercontent.com/u/72827868?v=4
+  username: andrei-kochin
+ayush.chaurarsia@gmail.com:
+  avatar: https://avatars.githubusercontent.com/u/15766192?v=4
+  username: AyushExel
+chr043416@gmail.com:
+  avatar: https://avatars.githubusercontent.com/u/62513924?v=4
+  username: RizwanMunawar
+glenn.jocher@ultralytics.com:
+  avatar: https://avatars.githubusercontent.com/u/26833433?v=4
+  username: glenn-jocher
+hnliu_2@stu.xidian.edu.cn:
+  avatar: null
+  username: null
+jpedrofonseca_94@hotmail.com:
+  avatar: null
+  username: null
+k-2feng@hotmail.com:
+  avatar: null
+  username: null
+lakshanthad@yahoo.com:
+  avatar: https://avatars.githubusercontent.com/u/20147381?v=4
+  username: lakshanthad
+muhammadrizwanmunawar123@gmail.com:
+  avatar: https://avatars.githubusercontent.com/u/62513924?v=4
+  username: RizwanMunawar
+plashchynski@gmail.com:
+  avatar: https://avatars.githubusercontent.com/u/30833?v=4
+  username: plashchynski
+priytosh.revolution@live.com:
+  avatar: https://avatars.githubusercontent.com/u/19519529?v=4
+  username: priytosh-tripathi
+rulosanti@gmail.com:
+  avatar: null
+  username: null
+shuizhuyuanluo@126.com:
+  avatar: null
+  username: null
+sometimesocrazy@gmail.com:
+  avatar: null
+  username: null
+stormsson@users.noreply.github.com:
+  avatar: https://avatars.githubusercontent.com/u/1133032?v=4
+  username: stormsson
+xinwang614@gmail.com:
+  avatar: https://avatars.githubusercontent.com/u/17264618?v=4
+  username: GreatV

examples/heatmaps.ipynb

@@ -116,7 +116,7 @@
     "    colormap=cv2.COLORMAP_PARULA,\n",
     "    view_img=True,\n",
     "    shape=\"circle\",\n",
-    "    classes_names=model.names,\n",
+    "    names=model.names,\n",
     ")\n",
     "\n",
     "while cap.isOpened():\n",

examples/object_counting.ipynb

@@ -129,7 +129,7 @@
     "counter = solutions.ObjectCounter(\n",
     "    view_img=True,  # Display the image during processing\n",
     "    reg_pts=line_points,  # Region of interest points\n",
-    "    classes_names=model.names,  # Class names from the YOLO model\n",
+    "    names=model.names,  # Class names from the YOLO model\n",
     "    draw_tracks=True,  # Draw tracking lines for objects\n",
     "    line_thickness=2,  # Thickness of the lines drawn\n",
     ")\n",

pyproject.toml

@@ -71,6 +71,7 @@ dependencies = [
     "pyyaml>=5.3.1",
     "requests>=2.23.0",
     "scipy>=1.4.1",
+    "torch>=1.8.0,<2.4.0; sys_platform == 'win32'",  # Windows CPU errors https://github.com/ultralytics/ultralytics/issues/15049
     "torch>=1.8.0",
     "torchvision>=0.9.0",
     "tqdm>=4.64.0", # progress bars
@@ -93,7 +94,7 @@ dev = [
     "mkdocstrings[python]",
     "mkdocs-jupyter", # notebooks
     "mkdocs-redirects", # 301 redirects
-    "mkdocs-ultralytics-plugin>=0.1.2", # for meta descriptions and images, dates and authors
+    "mkdocs-ultralytics-plugin>=0.1.6", # for meta descriptions and images, dates and authors
     "mkdocs-macros-plugin>=1.0.5"  # duplicating content (i.e. export tables) in multiple places
 ]
 export = [

tests/test_explorer.py

@@ -5,9 +5,11 @@ import pytest
 
 from ultralytics import Explorer
 from ultralytics.utils import ASSETS
+from ultralytics.utils.torch_utils import TORCH_1_13
 
 
 @pytest.mark.slow
+@pytest.mark.skipif(not TORCH_1_13, reason="Explorer requires torch>=1.13")
 def test_similarity():
     """Test the correctness and response length of similarity calculations and SQL queries in the Explorer."""
     exp = Explorer(data="coco8.yaml")
@@ -25,6 +27,7 @@ def test_similarity():
 
 
 @pytest.mark.slow
+@pytest.mark.skipif(not TORCH_1_13, reason="Explorer requires torch>=1.13")
 def test_det():
     """Test detection functionalities and verify embedding table includes bounding boxes."""
     exp = Explorer(data="coco8.yaml", model="yolov8n.pt")
@@ -38,6 +41,7 @@ def test_det():
 
 
 @pytest.mark.slow
+@pytest.mark.skipif(not TORCH_1_13, reason="Explorer requires torch>=1.13")
 def test_seg():
     """Test segmentation functionalities and ensure the embedding table includes segmentation masks."""
     exp = Explorer(data="coco8-seg.yaml", model="yolov8n-seg.pt")
@@ -50,6 +54,7 @@ def test_seg():
 
 
 @pytest.mark.slow
+@pytest.mark.skipif(not TORCH_1_13, reason="Explorer requires torch>=1.13")
 def test_pose():
     """Test pose estimation functionality and verify the embedding table includes keypoints."""
     exp = Explorer(data="coco8-pose.yaml", model="yolov8n-pose.pt")

tests/test_python.py

@@ -252,6 +252,8 @@ def test_labels_and_crops():
     for r in results:
         im_name = Path(r.path).stem
         cls_idxs = r.boxes.cls.int().tolist()
+        # Check correct detections
+        assert cls_idxs == ([0, 0, 5, 0, 7] if r.path.endswith("bus.jpg") else [0, 0])  # bus.jpg and zidane.jpg classes
         # Check label path
         labels = save_path / f"labels/{im_name}.txt"
         assert labels.exists()

ultralytics/__init__.py

@@ -1,6 +1,6 @@
 # Ultralytics YOLO 🚀, AGPL-3.0 license
 
-__version__ = "8.2.83"
+__version__ = "8.2.86"
 
 import os
 

ultralytics/data/converter.py

@@ -490,7 +490,7 @@ def convert_dota_to_yolo_obb(dota_root_path: str):
                 normalized_coords = [
                     coords[i] / image_width if i % 2 == 0 else coords[i] / image_height for i in range(8)
                 ]
-                formatted_coords = ["{:.6g}".format(coord) for coord in normalized_coords]
+                formatted_coords = [f"{coord:.6g}" for coord in normalized_coords]
                 g.write(f"{class_idx} {' '.join(formatted_coords)}\n")
 
     for phase in ["train", "val"]:

ultralytics/data/dataset.py

@@ -296,7 +296,7 @@ class GroundingDataset(YOLODataset):
         """Loads annotations from a JSON file, filters, and normalizes bounding boxes for each image."""
         labels = []
         LOGGER.info("Loading annotation file...")
-        with open(self.json_file, "r") as f:
+        with open(self.json_file) as f:
             annotations = json.load(f)
         images = {f'{x["id"]:d}': x for x in annotations["images"]}
         img_to_anns = defaultdict(list)

ultralytics/data/split_dota.py

@@ -193,7 +193,7 @@ def crop_and_save(anno, windows, window_objs, im_dir, lb_dir, allow_background_i
 
             with open(Path(lb_dir) / f"{new_name}.txt", "w") as f:
                 for lb in label:
-                    formatted_coords = ["{:.6g}".format(coord) for coord in lb[1:]]
+                    formatted_coords = [f"{coord:.6g}" for coord in lb[1:]]
                     f.write(f"{int(lb[0])} {' '.join(formatted_coords)}\n")
 
 

ultralytics/engine/exporter.py

@@ -138,7 +138,7 @@ def try_export(inner_func):
             LOGGER.info(f"{prefix} export success ✅ {dt.t:.1f}s, saved as '{f}' ({file_size(f):.1f} MB)")
             return f, model
         except Exception as e:
-            LOGGER.info(f"{prefix} export failure ❌ {dt.t:.1f}s: {e}")
+            LOGGER.error(f"{prefix} export failure ❌ {dt.t:.1f}s: {e}")
             raise e
 
     return outer_func
@@ -204,8 +204,8 @@ class Exporter:
             self.args.half = False
             assert not self.args.dynamic, "half=True not compatible with dynamic=True, i.e. use only one."
         self.imgsz = check_imgsz(self.args.imgsz, stride=model.stride, min_dim=2)  # check image size
-        if self.args.int8 and (engine or xml):
-            self.args.dynamic = True  # enforce dynamic to export TensorRT INT8; ensures ONNX is dynamic
+        if self.args.int8 and engine:
+            self.args.dynamic = True  # enforce dynamic to export TensorRT INT8
         if self.args.optimize:
             assert not ncnn, "optimize=True not compatible with format='ncnn', i.e. use optimize=False"
             assert self.device.type == "cpu", "optimize=True not compatible with cuda devices, i.e. use device='cpu'"
@@ -248,6 +248,7 @@ class Exporter:
                 m.dynamic = self.args.dynamic
                 m.export = True
                 m.format = self.args.format
+                m.max_det = self.args.max_det
             elif isinstance(m, C2f) and not is_tf_format:
                 # EdgeTPU does not support FlexSplitV while split provides cleaner ONNX graph
                 m.forward = m.forward_split
@@ -353,18 +354,20 @@ class Exporter:
         """Build and return a dataloader suitable for calibration of INT8 models."""
         LOGGER.info(f"{prefix} collecting INT8 calibration images from 'data={self.args.data}'")
         data = (check_cls_dataset if self.model.task == "classify" else check_det_dataset)(self.args.data)
+        # TensorRT INT8 calibration should use 2x batch size
+        batch = self.args.batch * (2 if self.args.format == "engine" else 1)
         dataset = YOLODataset(
             data[self.args.split or "val"],
             data=data,
             task=self.model.task,
             imgsz=self.imgsz[0],
             augment=False,
-            batch_size=self.args.batch * 2,  # NOTE TensorRT INT8 calibration should use 2x batch size
+            batch_size=batch,
         )
         n = len(dataset)
         if n < 300:
             LOGGER.warning(f"{prefix} WARNING ⚠️ >300 images recommended for INT8 calibration, found {n} images.")
-        return build_dataloader(dataset, batch=self.args.batch * 2, workers=0)  # required for batch loading
+        return build_dataloader(dataset, batch=batch, workers=0)  # required for batch loading
 
     @try_export
     def export_torchscript(self, prefix=colorstr("TorchScript:")):
@@ -420,7 +423,6 @@ class Exporter:
 
         # Checks
         model_onnx = onnx.load(f)  # load onnx model
-        # onnx.checker.check_model(model_onnx)  # check onnx model
 
         # Simplify
         if self.args.simplify:
@@ -430,10 +432,6 @@ class Exporter:
                 LOGGER.info(f"{prefix} slimming with onnxslim {onnxslim.__version__}...")
                 model_onnx = onnxslim.slim(model_onnx)
 
-                # ONNX Simplifier (deprecated as must be compiled with 'cmake' in aarch64 and Conda CI environments)
-                # import onnxsim
-                # model_onnx, check = onnxsim.simplify(model_onnx)
-                # assert check, "Simplified ONNX model could not be validated"
             except Exception as e:
                 LOGGER.warning(f"{prefix} simplifier failure: {e}")
 
@@ -677,7 +675,6 @@ class Exporter:
     def export_engine(self, prefix=colorstr("TensorRT:")):
         """YOLOv8 TensorRT export https://developer.nvidia.com/tensorrt."""
         assert self.im.device.type != "cpu", "export running on CPU but must be on GPU, i.e. use 'device=0'"
-        # self.args.simplify = True
         f_onnx, _ = self.export_onnx()  # run before TRT import https://github.com/ultralytics/ultralytics/issues/7016
 
         try:
@@ -784,7 +781,7 @@ class Exporter:
             # Load dataset w/ builder (for batching) and calibrate
             config.int8_calibrator = EngineCalibrator(
                 dataset=self.get_int8_calibration_dataloader(prefix),
-                batch=2 * self.args.batch,
+                batch=2 * self.args.batch,  # TensorRT INT8 calibration should use 2x batch size
                 cache=str(self.file.with_suffix(".cache")),
             )
 
@@ -867,8 +864,6 @@ class Exporter:
                 f.mkdir()
                 images = [batch["img"].permute(0, 2, 3, 1) for batch in self.get_int8_calibration_dataloader(prefix)]
                 images = torch.cat(images, 0).float()
-                # mean = images.view(-1, 3).mean(0)  # imagenet mean [123.675, 116.28, 103.53]
-                # std = images.view(-1, 3).std(0)  # imagenet std [58.395, 57.12, 57.375]
                 np.save(str(tmp_file), images.numpy().astype(np.float32))  # BHWC
                 np_data = [["images", tmp_file, [[[[0, 0, 0]]]], [[[[255, 255, 255]]]]]]
         else:
@@ -996,20 +991,7 @@ class Exporter:
         if " " in f:
             LOGGER.warning(f"{prefix} WARNING ⚠️ your model may not work correctly with spaces in path '{f}'.")
 
-        # f_json = Path(f) / 'model.json'  # *.json path
-        # with open(f_json, 'w') as j:  # sort JSON Identity_* in ascending order
-        #     subst = re.sub(
-        #         r'{"outputs": {"Identity.?.?": {"name": "Identity.?.?"}, '
-        #         r'"Identity.?.?": {"name": "Identity.?.?"}, '
-        #         r'"Identity.?.?": {"name": "Identity.?.?"}, '
-        #         r'"Identity.?.?": {"name": "Identity.?.?"}}}',
-        #         r'{"outputs": {"Identity": {"name": "Identity"}, '
-        #         r'"Identity_1": {"name": "Identity_1"}, '
-        #         r'"Identity_2": {"name": "Identity_2"}, '
-        #         r'"Identity_3": {"name": "Identity_3"}}}',
-        #         f_json.read_text(),
-        #     )
-        #     j.write(subst)
+        # Add metadata
         yaml_save(Path(f) / "metadata.yaml", self.metadata)  # add metadata.yaml
         return f, None
 
@@ -1102,27 +1084,11 @@ class Exporter:
         names = self.metadata["names"]
         nx, ny = spec.description.input[0].type.imageType.width, spec.description.input[0].type.imageType.height
         _, nc = out0_shape  # number of anchors, number of classes
-        # _, nc = out0.type.multiArrayType.shape
         assert len(names) == nc, f"{len(names)} names found for nc={nc}"  # check
 
         # Define output shapes (missing)
         out0.type.multiArrayType.shape[:] = out0_shape  # (3780, 80)
         out1.type.multiArrayType.shape[:] = out1_shape  # (3780, 4)
-        # spec.neuralNetwork.preprocessing[0].featureName = '0'
-
-        # Flexible input shapes
-        # from coremltools.models.neural_network import flexible_shape_utils
-        # s = [] # shapes
-        # s.append(flexible_shape_utils.NeuralNetworkImageSize(320, 192))
-        # s.append(flexible_shape_utils.NeuralNetworkImageSize(640, 384))  # (height, width)
-        # flexible_shape_utils.add_enumerated_image_sizes(spec, feature_name='image', sizes=s)
-        # r = flexible_shape_utils.NeuralNetworkImageSizeRange()  # shape ranges
-        # r.add_height_range((192, 640))
-        # r.add_width_range((192, 640))
-        # flexible_shape_utils.update_image_size_range(spec, feature_name='image', size_range=r)
-
-        # Print
-        # print(spec.description)
 
         # Model from spec
         model = ct.models.MLModel(spec, weights_dir=weights_dir)

ultralytics/engine/predictor.py

@@ -328,7 +328,7 @@ class BasePredictor:
             frame = int(match[1]) if match else None  # 0 if frame undetermined
 
         self.txt_path = self.save_dir / "labels" / (p.stem + ("" if self.dataset.mode == "image" else f"_{frame}"))
-        string += "%gx%g " % im.shape[2:]
+        string += "{:g}x{:g} ".format(*im.shape[2:])
         result = self.results[i]
         result.save_dir = self.save_dir.__str__()  # used in other locations
         string += f"{result.verbose()}{result.speed['inference']:.1f}ms"

ultralytics/engine/validator.py

@@ -202,8 +202,9 @@ class BaseValidator:
             return {k: round(float(v), 5) for k, v in results.items()}  # return results as 5 decimal place floats
         else:
             LOGGER.info(
-                "Speed: %.1fms preprocess, %.1fms inference, %.1fms loss, %.1fms postprocess per image"
-                % tuple(self.speed.values())
+                "Speed: {:.1f}ms preprocess, {:.1f}ms inference, {:.1f}ms loss, {:.1f}ms postprocess per image".format(
+                    *tuple(self.speed.values())
+                )
             )
             if self.args.save_json and self.jdict:
                 with open(str(self.save_dir / "predictions.json"), "w") as f:

ultralytics/hub/utils.py

@@ -55,23 +55,22 @@ def request_with_credentials(url: str) -> any:
 
     display.display(
         display.Javascript(
-            """
-            window._hub_tmp = new Promise((resolve, reject) => {
+            f"""
+            window._hub_tmp = new Promise((resolve, reject) => {{
                 const timeout = setTimeout(() => reject("Failed authenticating existing browser session"), 5000)
-                fetch("%s", {
+                fetch("{url}", {{
                     method: 'POST',
                     credentials: 'include'
-                })
+                }})
                     .then((response) => resolve(response.json()))
-                    .then((json) => {
+                    .then((json) => {{
                     clearTimeout(timeout);
-                    }).catch((err) => {
+                    }}).catch((err) => {{
                     clearTimeout(timeout);
                     reject(err);
-                });
-            });
+                }});
+            }});
             """
-            % url
         )
     )
     return output.eval_js("_hub_tmp")

ultralytics/models/fastsam/predict.py

@@ -100,7 +100,7 @@ class FastSAMPredictor(SegmentationPredictor):
                     texts = [texts]
                 crop_ims, filter_idx = [], []
                 for i, b in enumerate(result.boxes.xyxy.tolist()):
-                    x1, y1, x2, y2 = [int(x) for x in b]
+                    x1, y1, x2, y2 = (int(x) for x in b)
                     if masks[i].sum() <= 100:
                         filter_idx.append(i)
                         continue

ultralytics/models/sam/model.py

@@ -106,7 +106,7 @@ class SAM(Model):
             ...     print(f"Detected {len(r.masks)} masks")
         """
         overrides = dict(conf=0.25, task="segment", mode="predict", imgsz=1024)
-        kwargs.update(overrides)
+        kwargs = {**overrides, **kwargs}
         prompts = dict(bboxes=bboxes, points=points, labels=labels)
         return super().predict(source, stream, prompts=prompts, **kwargs)
 

ultralytics/models/sam/modules/blocks.py

@@ -35,7 +35,7 @@ class DropPath(nn.Module):
 
     def __init__(self, drop_prob=0.0, scale_by_keep=True):
         """Initialize DropPath module for stochastic depth regularization during training."""
-        super(DropPath, self).__init__()
+        super().__init__()
         self.drop_prob = drop_prob
         self.scale_by_keep = scale_by_keep
 

ultralytics/models/sam/modules/encoders.py

@@ -151,7 +151,12 @@ class ImageEncoderViT(nn.Module):
         """Processes input through patch embedding, positional embedding, transformer blocks, and neck module."""
         x = self.patch_embed(x)
         if self.pos_embed is not None:
-            x = x + self.pos_embed
+            pos_embed = (
+                F.interpolate(self.pos_embed.permute(0, 3, 1, 2), scale_factor=self.img_size / 1024).permute(0, 2, 3, 1)
+                if self.img_size != 1024
+                else self.pos_embed
+            )
+            x = x + pos_embed
         for blk in self.blocks:
             x = blk(x)
         return self.neck(x.permute(0, 3, 1, 2))

ultralytics/models/sam/modules/sam.py

@@ -90,6 +90,19 @@ class SAMModel(nn.Module):
         self.register_buffer("pixel_mean", torch.Tensor(pixel_mean).view(-1, 1, 1), False)
         self.register_buffer("pixel_std", torch.Tensor(pixel_std).view(-1, 1, 1), False)
 
+    def set_imgsz(self, imgsz):
+        """
+        Set image size to make model compatible with different image sizes.
+
+        Args:
+            imgsz (Tuple[int, int]): The size of the input image.
+        """
+        if hasattr(self.image_encoder, "set_imgsz"):
+            self.image_encoder.set_imgsz(imgsz)
+        self.prompt_encoder.input_image_size = imgsz
+        self.prompt_encoder.image_embedding_size = [x // 16 for x in imgsz]  # 16 is fixed as patch size of ViT model
+        self.image_encoder.img_size = imgsz[0]
+
 
 class SAM2Model(torch.nn.Module):
     """
@@ -940,3 +953,14 @@ class SAM2Model(torch.nn.Module):
         # don't overlap (here sigmoid(-10.0)=4.5398e-05)
         pred_masks = torch.where(keep, pred_masks, torch.clamp(pred_masks, max=-10.0))
         return pred_masks
+
+    def set_imgsz(self, imgsz):
+        """
+        Set image size to make model compatible with different image sizes.
+
+        Args:
+            imgsz (Tuple[int, int]): The size of the input image.
+        """
+        self.image_size = imgsz[0]
+        self.sam_prompt_encoder.input_image_size = imgsz
+        self.sam_prompt_encoder.image_embedding_size = [x // 16 for x in imgsz]  # fixed ViT patch size of 16

ultralytics/models/sam/modules/tiny_encoder.py

@@ -982,10 +982,31 @@ class TinyViT(nn.Module):
             layer = self.layers[i]
             x = layer(x)
         batch, _, channel = x.shape
-        x = x.view(batch, 64, 64, channel)
+        x = x.view(batch, self.patches_resolution[0] // 4, self.patches_resolution[1] // 4, channel)
         x = x.permute(0, 3, 1, 2)
         return self.neck(x)
 
     def forward(self, x):
         """Performs the forward pass through the TinyViT model, extracting features from the input image."""
         return self.forward_features(x)
+
+    def set_imgsz(self, imgsz=[1024, 1024]):
+        """
+        Set image size to make model compatible with different image sizes.
+
+        Args:
+            imgsz (Tuple[int, int]): The size of the input image.
+        """
+        imgsz = [s // 4 for s in imgsz]
+        self.patches_resolution = imgsz
+        for i, layer in enumerate(self.layers):
+            input_resolution = (
+                imgsz[0] // (2 ** (i - 1 if i == 3 else i)),
+                imgsz[1] // (2 ** (i - 1 if i == 3 else i)),
+            )
+            layer.input_resolution = input_resolution
+            if layer.downsample is not None:
+                layer.downsample.input_resolution = input_resolution
+            if isinstance(layer, BasicLayer):
+                for b in layer.blocks:
+                    b.input_resolution = input_resolution

ultralytics/models/sam/predict.py

@@ -95,7 +95,7 @@ class Predictor(BasePredictor):
         """
         if overrides is None:
             overrides = {}
-        overrides.update(dict(task="segment", mode="predict", imgsz=1024))
+        overrides.update(dict(task="segment", mode="predict"))
         super().__init__(cfg, overrides, _callbacks)
         self.args.retina_masks = True
         self.im = None
@@ -455,8 +455,11 @@ class Predictor(BasePredictor):
                 cls = torch.arange(len(pred_masks), dtype=torch.int32, device=pred_masks.device)
                 pred_bboxes = torch.cat([pred_bboxes, pred_scores[:, None], cls[:, None]], dim=-1)
 
-            masks = ops.scale_masks(masks[None].float(), orig_img.shape[:2], padding=False)[0]
-            masks = masks > self.model.mask_threshold  # to bool
+            if len(masks) == 0:
+                masks = None
+            else:
+                masks = ops.scale_masks(masks[None].float(), orig_img.shape[:2], padding=False)[0]
+                masks = masks > self.model.mask_threshold  # to bool
             results.append(Results(orig_img, path=img_path, names=names, masks=masks, boxes=pred_bboxes))
         # Reset segment-all mode.
         self.segment_all = False
@@ -522,6 +525,10 @@ class Predictor(BasePredictor):
 
     def get_im_features(self, im):
         """Extracts image features using the SAM model's image encoder for subsequent mask prediction."""
+        assert (
+            isinstance(self.imgsz, (tuple, list)) and self.imgsz[0] == self.imgsz[1]
+        ), f"SAM models only support square image size, but got {self.imgsz}."
+        self.model.set_imgsz(self.imgsz)
         return self.model.image_encoder(im)
 
     def set_prompts(self, prompts):
@@ -761,6 +768,12 @@ class SAM2Predictor(Predictor):
 
     def get_im_features(self, im):
         """Extracts image features from the SAM image encoder for subsequent processing."""
+        assert (
+            isinstance(self.imgsz, (tuple, list)) and self.imgsz[0] == self.imgsz[1]
+        ), f"SAM 2 models only support square image size, but got {self.imgsz}."
+        self.model.set_imgsz(self.imgsz)
+        self._bb_feat_sizes = [[x // (4 * i) for x in self.imgsz] for i in [1, 2, 4]]
+
         backbone_out = self.model.forward_image(im)
         _, vision_feats, _, _ = self.model._prepare_backbone_features(backbone_out)
         if self.model.directly_add_no_mem_embed:

ultralytics/nn/modules/block.py

@@ -672,7 +672,7 @@ class CBLinear(nn.Module):
 
     def __init__(self, c1, c2s, k=1, s=1, p=None, g=1):
         """Initializes the CBLinear module, passing inputs unchanged."""
-        super(CBLinear, self).__init__()
+        super().__init__()
         self.c2s = c2s
         self.conv = nn.Conv2d(c1, sum(c2s), k, s, autopad(k, p), groups=g, bias=True)
 
@@ -686,7 +686,7 @@ class CBFuse(nn.Module):
 
     def __init__(self, idx):
         """Initializes CBFuse module with layer index for selective feature fusion."""
-        super(CBFuse, self).__init__()
+        super().__init__()
         self.idx = idx
 
     def forward(self, xs):

ultralytics/nn/modules/head.py

@@ -144,12 +144,12 @@ class Detect(nn.Module):
             (torch.Tensor): Processed predictions with shape (batch_size, min(max_det, num_anchors), 6) and last
                 dimension format [x, y, w, h, max_class_prob, class_index].
         """
-        batch_size, anchors, predictions = preds.shape  # i.e. shape(16,8400,84)
+        batch_size, anchors, _ = preds.shape  # i.e. shape(16,8400,84)
         boxes, scores = preds.split([4, nc], dim=-1)
         index = scores.amax(dim=-1).topk(min(max_det, anchors))[1].unsqueeze(-1)
         boxes = boxes.gather(dim=1, index=index.repeat(1, 1, 4))
         scores = scores.gather(dim=1, index=index.repeat(1, 1, nc))
-        scores, index = scores.flatten(1).topk(max_det)
+        scores, index = scores.flatten(1).topk(min(max_det, anchors))
         i = torch.arange(batch_size)[..., None]  # batch indices
         return torch.cat([boxes[i, index // nc], scores[..., None], (index % nc)[..., None].float()], dim=-1)
 

ultralytics/solutions/parking_management.py

@@ -210,7 +210,7 @@ class ParkingManagement:
         Args:
             json_file (str): file that have all parking slot points
         """
-        with open(json_file, "r") as f:
+        with open(json_file) as f:
             return json.load(f)
 
     def process_data(self, json_data, im0, boxes, clss):

ultralytics/utils/benchmarks.py

@@ -213,7 +213,7 @@ class RF100Benchmark:
         os.mkdir("ultralytics-benchmarks")
         safe_download("https://github.com/ultralytics/assets/releases/download/v0.0.0/datasets_links.txt")
 
-        with open(ds_link_txt, "r") as file:
+        with open(ds_link_txt) as file:
             for line in file:
                 try:
                     _, url, workspace, project, version = re.split("/+", line.strip())
@@ -237,7 +237,7 @@ class RF100Benchmark:
         Args:
             path (str): YAML file path.
         """
-        with open(path, "r") as file:
+        with open(path) as file:
             yaml_data = yaml.safe_load(file)
         yaml_data["train"] = "train/images"
         yaml_data["val"] = "valid/images"
@@ -257,7 +257,7 @@ class RF100Benchmark:
         skip_symbols = ["🚀", "⚠️", "💡", "❌"]
         with open(yaml_path) as stream:
             class_names = yaml.safe_load(stream)["names"]
-        with open(val_log_file, "r", encoding="utf-8") as f:
+        with open(val_log_file, encoding="utf-8") as f:
             lines = f.readlines()
             eval_lines = []
             for line in lines:

ultralytics/utils/checks.py

@@ -29,11 +29,13 @@ from ultralytics.utils import (
     IS_PIP_PACKAGE,
     LINUX,
     LOGGER,
+    MACOS,
     ONLINE,
     PYTHON_VERSION,
     ROOT,
     TORCHVISION_VERSION,
     USER_CONFIG_DIR,
+    WINDOWS,
     Retry,
     SimpleNamespace,
     ThreadingLocked,
@@ -224,6 +226,14 @@ def check_version(
     if not required:  # if required is '' or None
         return True
 
+    if "sys_platform" in required:  # i.e. required='<2.4.0,>=1.8.0; sys_platform == "win32"'
+        if (
+            (WINDOWS and "win32" not in required)
+            or (LINUX and "linux" not in required)
+            or (MACOS and "macos" not in required and "darwin" not in required)
+        ):
+            return True
+
     op = ""
     version = ""
     result = True
@@ -422,6 +432,7 @@ def check_torchvision():
     """
     # Compatibility table
     compatibility_table = {
+        "2.4": ["0.19"],
         "2.3": ["0.18"],
         "2.2": ["0.17"],
         "2.1": ["0.16"],

ultralytics/utils/metrics.py

@@ -460,7 +460,7 @@ def plot_pr_curve(px, py, ap, save_dir=Path("pr_curve.png"), names={}, on_plot=N
     else:
         ax.plot(px, py, linewidth=1, color="grey")  # plot(recall, precision)
 
-    ax.plot(px, py.mean(1), linewidth=3, color="blue", label="all classes %.3f mAP@0.5" % ap[:, 0].mean())
+    ax.plot(px, py.mean(1), linewidth=3, color="blue", label=f"all classes {ap[:, 0].mean():.3f} mAP@0.5")
     ax.set_xlabel("Recall")
     ax.set_ylabel("Precision")
     ax.set_xlim(0, 1)

ultralytics/utils/ops.py

@@ -218,7 +218,7 @@ def non_max_suppression(
         classes = torch.tensor(classes, device=prediction.device)
 
     if prediction.shape[-1] == 6:  # end-to-end model (BNC, i.e. 1,300,6)
-        output = [pred[pred[:, 4] > conf_thres] for pred in prediction]
+        output = [pred[pred[:, 4] > conf_thres][:max_det] for pred in prediction]
         if classes is not None:
             output = [pred[(pred[:, 5:6] == classes).any(1)] for pred in output]
         return output

ultralytics/utils/torch_utils.py

@@ -1,4 +1,5 @@
 # Ultralytics YOLO 🚀, AGPL-3.0 license
+
 import contextlib
 import gc
 import math
@@ -24,6 +25,7 @@ from ultralytics.utils import (
     NUM_THREADS,
     PYTHON_VERSION,
     TORCHVISION_VERSION,
+    WINDOWS,
     __version__,
     colorstr,
 )
@@ -42,6 +44,11 @@ TORCHVISION_0_10 = check_version(TORCHVISION_VERSION, "0.10.0")
 TORCHVISION_0_11 = check_version(TORCHVISION_VERSION, "0.11.0")
 TORCHVISION_0_13 = check_version(TORCHVISION_VERSION, "0.13.0")
 TORCHVISION_0_18 = check_version(TORCHVISION_VERSION, "0.18.0")
+if WINDOWS and torch.__version__[:3] == "2.4":  # reject all versions of 2.4 on Windows
+    LOGGER.warning(
+        "WARNING ⚠️ Known issue with torch>=2.4.0 on Windows with CPU, recommend downgrading to torch<=2.3.1 to resolve "
+        "https://github.com/ultralytics/ultralytics/issues/15049"
+    )
 
 
 @contextmanager