Add benchmarks to Docker publish workflow (#3931)

.github/workflows/ci.yaml

@@ -109,25 +109,17 @@ jobs:
           pip --version
           pip list
       - name: Benchmark DetectionModel
-        shell: python
-        run: |
-          from ultralytics.utils.benchmarks import benchmark
-          benchmark(model='path with spaces/${{ matrix.model }}.pt', imgsz=160, half=False, hard_fail=0.26)
+        shell: bash
+        run: yolo benchmark model='path with spaces/${{ matrix.model }}.pt' imgsz=160, verbose=0.26
       - name: Benchmark SegmentationModel
-        shell: python
-        run: |
-          from ultralytics.utils.benchmarks import benchmark
-          benchmark(model='path with spaces/${{ matrix.model }}-seg.pt', imgsz=160, half=False, hard_fail=0.30)
+        shell: bash
+        run: yolo benchmark model='path with spaces/${{ matrix.model }}-seg.pt' imgsz=160, verbose=0.30
       - name: Benchmark ClassificationModel
-        shell: python
-        run: |
-          from ultralytics.utils.benchmarks import benchmark
-          benchmark(model='path with spaces/${{ matrix.model }}-cls.pt', imgsz=160, half=False, hard_fail=0.36)
+        shell: bash
+        run: yolo benchmark model='path with spaces/${{ matrix.model }}-cls.pt' imgsz=160, verbose=0.36
       - name: Benchmark PoseModel
-        shell: python
-        run: |
-          from ultralytics.utils.benchmarks import benchmark
-          benchmark(model='path with spaces/${{ matrix.model }}-pose.pt', imgsz=160, half=False, hard_fail=0.17)
+        shell: bash
+        run: yolo benchmark model='path with spaces/${{ matrix.model }}-pose.pt' imgsz=160, verbose=0.17
       - name: Benchmark Summary
         run: |
           cat benchmarks.log

.github/workflows/docker.yaml

@@ -76,7 +76,7 @@ jobs:
       - name: Run Benchmarks
         if: matrix.platforms == 'linux/amd64'  # arm64 images not supported on GitHub CI runners
         run: |
-          docker run ultralytics/ultralytics:${{ matrix.tags }} yolo benchmark model=yolov8n.pt imgsz=160
+          docker run ultralytics/ultralytics:${{ matrix.tags }} yolo benchmark model=yolov8n.pt imgsz=160 verbose=0.26
 
       - name: Push Image
         if: github.event_name == 'push' || github.event.inputs.push == true

docker/Dockerfile

@@ -30,7 +30,15 @@ ADD https://github.com/ultralytics/assets/releases/download/v0.0.0/yolov8n.pt /u
 
 # Install pip packages
 RUN python3 -m pip install --upgrade pip wheel
-RUN pip install --no-cache -e . albumentations comet thop pycocotools onnx onnx-simplifier onnxruntime-gpu
+RUN pip install --no-cache -e '.[export]' thop albumentations comet pycocotools
+
+# Run exports to AutoInstall packages
+RUN yolo export model=tmp/yolov8n.pt format=edgetpu imgsz=32
+RUN yolo export model=tmp/yolov8n.pt format=ncnn imgsz=32
+# Requires <= Python 3.10, bug with paddlepaddle==2.5.0
+RUN pip install --no-cache paddlepaddle==2.4.2 x2paddle
+# Remove exported models
+RUN rm -rf tmp
 
 # Set environment variables
 ENV OMP_NUM_THREADS=1

docker/Dockerfile-cpu

@@ -28,8 +28,15 @@ RUN rm -rf /usr/lib/python3.11/EXTERNALLY-MANAGED
 
 # Install pip packages
 RUN python3 -m pip install --upgrade pip wheel
-RUN pip install --no-cache -e . thop --extra-index-url https://download.pytorch.org/whl/cpu
-
+RUN pip install --no-cache -e '.[export]' thop --extra-index-url https://download.pytorch.org/whl/cpu
+
+# Run exports to AutoInstall packages
+RUN yolo export model=tmp/yolov8n.pt format=edgetpu imgsz=32
+RUN yolo export model=tmp/yolov8n.pt format=ncnn imgsz=32
+# Requires <= Python 3.10, bug with paddlepaddle==2.5.0
+# RUN pip install --no-cache paddlepaddle==2.4.2 x2paddle
+# Remove exported models
+RUN rm -rf tmp
 
 # Usage Examples -------------------------------------------------------------------------------------------------------
 

docker/Dockerfile-python

@@ -31,15 +31,12 @@ RUN python3 -m pip install --upgrade pip wheel
 RUN pip install --no-cache -e '.[export]' thop --extra-index-url https://download.pytorch.org/whl/cpu
 
 # Run exports to AutoInstall packages
-WORKDIR /tmp_exports
-RUN yolo export format=edgetpu imgsz=32
-RUN yolo export format=ncnn imgsz=32
+RUN yolo export model=tmp/yolov8n.pt format=edgetpu imgsz=32
+RUN yolo export model=tmp/yolov8n.pt format=ncnn imgsz=32
 # Requires <= Python 3.10, bug with paddlepaddle==2.5.0
 RUN pip install --no-cache paddlepaddle==2.4.2 x2paddle
-
-# Reset workdir
-WORKDIR /usr/src/ultralytics
-RUN rm -rf /tmp_exports
+# Remove exported models
+RUN rm -rf tmp
 
 # Usage Examples -------------------------------------------------------------------------------------------------------
 

docs/modes/benchmark.md

@@ -40,18 +40,18 @@ full list of export arguments.
 
 ## Arguments
 
-Arguments such as `model`, `data`, `imgsz`, `half`, `device`, and `hard_fail` provide users with the flexibility to fine-tune
+Arguments such as `model`, `data`, `imgsz`, `half`, `device`, and `verbose` provide users with the flexibility to fine-tune
 the benchmarks to their specific needs and compare the performance of different export formats with ease.
 
-| Key         | Value   | Description                                                                |
-|-------------|---------|----------------------------------------------------------------------------|
-| `model`     | `None`  | path to model file, i.e. yolov8n.pt, yolov8n.yaml                          |
-| `data`      | `None`  | path to yaml referencing the benchmarking dataset (under `val` label)      |
-| `imgsz`     | `640`   | image size as scalar or (h, w) list, i.e. (640, 480)                       |
-| `half`      | `False` | FP16 quantization                                                          |
-| `int8`      | `False` | INT8 quantization                                                          |
-| `device`    | `None`  | device to run on, i.e. cuda device=0 or device=0,1,2,3 or device=cpu       |
-| `hard_fail` | `False` | do not continue on error (bool), or val floor threshold (float)            |
+| Key       | Value   | Description                                                           |
+|-----------|---------|-----------------------------------------------------------------------|
+| `model`   | `None`  | path to model file, i.e. yolov8n.pt, yolov8n.yaml                     |
+| `data`    | `None`  | path to yaml referencing the benchmarking dataset (under `val` label) |
+| `imgsz`   | `640`   | image size as scalar or (h, w) list, i.e. (640, 480)                  |
+| `half`    | `False` | FP16 quantization                                                     |
+| `int8`    | `False` | INT8 quantization                                                     |
+| `device`  | `None`  | device to run on, i.e. cuda device=0 or device=0,1,2,3 or device=cpu  |
+| `verbose` | `False` | do not continue on error (bool), or val floor threshold (float)       |
 
 ## Export Formats
 

ultralytics/engine/model.py

@@ -319,7 +319,11 @@ class YOLO:
         overrides.update(kwargs)
         overrides['mode'] = 'benchmark'
         overrides = {**DEFAULT_CFG_DICT, **overrides}  # fill in missing overrides keys with defaults
-        return benchmark(model=self, imgsz=overrides['imgsz'], half=overrides['half'], device=overrides['device'])
+        return benchmark(model=self,
+                         imgsz=overrides['imgsz'],
+                         half=overrides['half'],
+                         device=overrides['device'],
+                         verbose=overrides['verbose'])
 
     def export(self, **kwargs):
         """

ultralytics/utils/benchmarks.py

@@ -26,6 +26,7 @@ ncnn                    | `ncnn`                    | yolov8n_ncnn_model/
 
 import glob
 import platform
+import sys
 import time
 from pathlib import Path
 
@@ -49,7 +50,7 @@ def benchmark(model=Path(SETTINGS['weights_dir']) / 'yolov8n.pt',
               half=False,
               int8=False,
               device='cpu',
-              hard_fail=False):
+              verbose=False):
     """
     Benchmark a YOLO model across different formats for speed and accuracy.
 
@@ -61,7 +62,7 @@ def benchmark(model=Path(SETTINGS['weights_dir']) / 'yolov8n.pt',
         half (bool, optional): Use half-precision for the model if True. Default is False.
         int8 (bool, optional): Use int8-precision for the model if True. Default is False.
         device (str, optional): Device to run the benchmark on, either 'cpu' or 'cuda'. Default is 'cpu'.
-        hard_fail (bool | float | optional): If True or a float, assert benchmarks pass with given metric.
+        verbose (bool | float | optional): If True or a float, assert benchmarks pass with given metric.
             Default is False.
 
     Returns:
@@ -84,6 +85,8 @@ def benchmark(model=Path(SETTINGS['weights_dir']) / 'yolov8n.pt',
             assert i != 9 or LINUX, 'Edge TPU export only supported on Linux'
             if i == 10:
                 assert MACOS or LINUX, 'TF.js export only supported on macOS and Linux'
+            elif i == 11:
+                assert sys.version_info < (3, 11), 'PaddlePaddle export only supported on Python<=3.10'
             if 'cpu' in device.type:
                 assert cpu, 'inference not supported on CPU'
             if 'cuda' in device.type:
@@ -121,7 +124,7 @@ def benchmark(model=Path(SETTINGS['weights_dir']) / 'yolov8n.pt',
             metric, speed = results.results_dict[key], results.speed['inference']
             y.append([name, '✅', round(file_size(filename), 1), round(metric, 4), round(speed, 2)])
         except Exception as e:
-            if hard_fail:
+            if verbose:
                 assert type(e) is AssertionError, f'Benchmark failure for {name}: {e}'
             LOGGER.warning(f'ERROR ❌️ Benchmark failure for {name}: {e}')
             y.append([name, emoji, round(file_size(filename), 1), None, None])  # mAP, t_inference
@@ -136,9 +139,9 @@ def benchmark(model=Path(SETTINGS['weights_dir']) / 'yolov8n.pt',
     with open('benchmarks.log', 'a', errors='ignore', encoding='utf-8') as f:
         f.write(s)
 
-    if hard_fail and isinstance(hard_fail, float):
+    if verbose and isinstance(verbose, float):
         metrics = df[key].array  # values to compare to floor
-        floor = hard_fail  # minimum metric floor to pass, i.e. = 0.29 mAP for YOLOv5n
+        floor = verbose  # minimum metric floor to pass, i.e. = 0.29 mAP for YOLOv5n
         assert all(x > floor for x in metrics if pd.notna(x)), f'Benchmark failure: metric(s) < floor {floor}'
 
     return df

ultralytics/utils/torch_utils.py

@@ -28,7 +28,7 @@ TORCHVISION_0_10 = check_version(torchvision.__version__, '0.10.0')
 TORCH_1_9 = check_version(torch.__version__, '1.9.0')
 TORCH_1_11 = check_version(torch.__version__, '1.11.0')
 TORCH_1_12 = check_version(torch.__version__, '1.12.0')
-TORCH_2_0 = check_version(torch.__version__, minimum='2.0')
+TORCH_2_0 = check_version(torch.__version__, '2.0.0')
 
 
 @contextmanager