Compare commits
19 Commits
release_co
...
main
| Author | SHA1 | Date |
|---|---|---|
Shilong Liu
|
9dac4c605b | 2 years ago |
SlongLiu
|
3bb2c86c9a | 2 years ago |
|
SlongLiu
|
d3bc35fdea | 2 years ago |
|
SlongLiu
|
15ade007a8 | 2 years ago |
|
Shilong Liu
|
22292c4b78 | 2 years ago |
rentainhe
|
4c8f9206b6 | 2 years ago |
|
rentainhe
|
97ad9935ac | 2 years ago |
George Pearse
|
e93548c805 | 2 years ago |
Piotr Skalski
|
e45c11c4c3 | 2 years ago |
|
Piotr Skalski
|
f6b1145481 | 2 years ago |
|
SlongLiu
|
3023d1a26f | 2 years ago |
|
SlongLiu
|
a02cf79301 | 2 years ago |
|
SlongLiu
|
67a3c1940d | 2 years ago |
|
SlongLiu
|
ac00bd4a36 | 2 years ago |
|
Piotr Skalski
|
c974f60d73 | 2 years ago |
|
SlongLiu
|
858efccbad | 2 years ago |
|
Piotr Skalski
|
2309f9f468 | 2 years ago |
|
SlongLiu
|
12ef464f9e | 2 years ago |
|
Shilong Liu
|
3e7a8ca2dc | 2 years ago |
15 changed files with 1863 additions and 72 deletions
@ -0,0 +1,125 @@ |
|||||||
|
import argparse |
||||||
|
from functools import partial |
||||||
|
import cv2 |
||||||
|
import requests |
||||||
|
import os |
||||||
|
from io import BytesIO |
||||||
|
from PIL import Image |
||||||
|
import numpy as np |
||||||
|
from pathlib import Path |
||||||
|
|
||||||
|
|
||||||
|
import warnings |
||||||
|
|
||||||
|
import torch |
||||||
|
|
||||||
|
# prepare the environment |
||||||
|
os.system("python setup.py build develop --user") |
||||||
|
os.system("pip install packaging==21.3") |
||||||
|
os.system("pip install gradio") |
||||||
|
|
||||||
|
|
||||||
|
warnings.filterwarnings("ignore") |
||||||
|
|
||||||
|
import gradio as gr |
||||||
|
|
||||||
|
from groundingdino.models import build_model |
||||||
|
from groundingdino.util.slconfig import SLConfig |
||||||
|
from groundingdino.util.utils import clean_state_dict |
||||||
|
from groundingdino.util.inference import annotate, load_image, predict |
||||||
|
import groundingdino.datasets.transforms as T |
||||||
|
|
||||||
|
from huggingface_hub import hf_hub_download |
||||||
|
|
||||||
|
|
||||||
|
|
||||||
|
# Use this command for evaluate the Grounding DINO model |
||||||
|
config_file = "groundingdino/config/GroundingDINO_SwinT_OGC.py" |
||||||
|
ckpt_repo_id = "ShilongLiu/GroundingDINO" |
||||||
|
ckpt_filenmae = "groundingdino_swint_ogc.pth" |
||||||
|
|
||||||
|
|
||||||
|
def load_model_hf(model_config_path, repo_id, filename, device='cpu'): |
||||||
|
args = SLConfig.fromfile(model_config_path) |
||||||
|
model = build_model(args) |
||||||
|
args.device = device |
||||||
|
|
||||||
|
cache_file = hf_hub_download(repo_id=repo_id, filename=filename) |
||||||
|
checkpoint = torch.load(cache_file, map_location='cpu') |
||||||
|
log = model.load_state_dict(clean_state_dict(checkpoint['model']), strict=False) |
||||||
|
print("Model loaded from {} \n => {}".format(cache_file, log)) |
||||||
|
_ = model.eval() |
||||||
|
return model |
||||||
|
|
||||||
|
def image_transform_grounding(init_image): |
||||||
|
transform = T.Compose([ |
||||||
|
T.RandomResize([800], max_size=1333), |
||||||
|
T.ToTensor(), |
||||||
|
T.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]) |
||||||
|
]) |
||||||
|
image, _ = transform(init_image, None) # 3, h, w |
||||||
|
return init_image, image |
||||||
|
|
||||||
|
def image_transform_grounding_for_vis(init_image): |
||||||
|
transform = T.Compose([ |
||||||
|
T.RandomResize([800], max_size=1333), |
||||||
|
]) |
||||||
|
image, _ = transform(init_image, None) # 3, h, w |
||||||
|
return image |
||||||
|
|
||||||
|
model = load_model_hf(config_file, ckpt_repo_id, ckpt_filenmae) |
||||||
|
|
||||||
|
def run_grounding(input_image, grounding_caption, box_threshold, text_threshold): |
||||||
|
init_image = input_image.convert("RGB") |
||||||
|
original_size = init_image.size |
||||||
|
|
||||||
|
_, image_tensor = image_transform_grounding(init_image) |
||||||
|
image_pil: Image = image_transform_grounding_for_vis(init_image) |
||||||
|
|
||||||
|
# run grounidng |
||||||
|
boxes, logits, phrases = predict(model, image_tensor, grounding_caption, box_threshold, text_threshold, device='cpu') |
||||||
|
annotated_frame = annotate(image_source=np.asarray(image_pil), boxes=boxes, logits=logits, phrases=phrases) |
||||||
|
image_with_box = Image.fromarray(cv2.cvtColor(annotated_frame, cv2.COLOR_BGR2RGB)) |
||||||
|
|
||||||
|
|
||||||
|
return image_with_box |
||||||
|
|
||||||
|
if __name__ == "__main__": |
||||||
|
|
||||||
|
parser = argparse.ArgumentParser("Grounding DINO demo", add_help=True) |
||||||
|
parser.add_argument("--debug", action="store_true", help="using debug mode") |
||||||
|
parser.add_argument("--share", action="store_true", help="share the app") |
||||||
|
args = parser.parse_args() |
||||||
|
|
||||||
|
block = gr.Blocks().queue() |
||||||
|
with block: |
||||||
|
gr.Markdown("# [Grounding DINO](https://github.com/IDEA-Research/GroundingDINO)") |
||||||
|
gr.Markdown("### Open-World Detection with Grounding DINO") |
||||||
|
|
||||||
|
with gr.Row(): |
||||||
|
with gr.Column(): |
||||||
|
input_image = gr.Image(source='upload', type="pil") |
||||||
|
grounding_caption = gr.Textbox(label="Detection Prompt") |
||||||
|
run_button = gr.Button(label="Run") |
||||||
|
with gr.Accordion("Advanced options", open=False): |
||||||
|
box_threshold = gr.Slider( |
||||||
|
label="Box Threshold", minimum=0.0, maximum=1.0, value=0.25, step=0.001 |
||||||
|
) |
||||||
|
text_threshold = gr.Slider( |
||||||
|
label="Text Threshold", minimum=0.0, maximum=1.0, value=0.25, step=0.001 |
||||||
|
) |
||||||
|
|
||||||
|
with gr.Column(): |
||||||
|
gallery = gr.outputs.Image( |
||||||
|
type="pil", |
||||||
|
# label="grounding results" |
||||||
|
).style(full_width=True, full_height=True) |
||||||
|
# gallery = gr.Gallery(label="Generated images", show_label=False).style( |
||||||
|
# grid=[1], height="auto", container=True, full_width=True, full_height=True) |
||||||
|
|
||||||
|
run_button.click(fn=run_grounding, inputs=[ |
||||||
|
input_image, grounding_caption, box_threshold, text_threshold], outputs=[gallery]) |
||||||
|
|
||||||
|
|
||||||
|
block.launch(server_name='0.0.0.0', server_port=7579, debug=args.debug, share=args.share) |
||||||
|
|
||||||
File diff suppressed because one or more lines are too long
File diff suppressed because one or more lines are too long
@ -0,0 +1,43 @@ |
|||||||
|
batch_size = 1 |
||||||
|
modelname = "groundingdino" |
||||||
|
backbone = "swin_B_384_22k" |
||||||
|
position_embedding = "sine" |
||||||
|
pe_temperatureH = 20 |
||||||
|
pe_temperatureW = 20 |
||||||
|
return_interm_indices = [1, 2, 3] |
||||||
|
backbone_freeze_keywords = None |
||||||
|
enc_layers = 6 |
||||||
|
dec_layers = 6 |
||||||
|
pre_norm = False |
||||||
|
dim_feedforward = 2048 |
||||||
|
hidden_dim = 256 |
||||||
|
dropout = 0.0 |
||||||
|
nheads = 8 |
||||||
|
num_queries = 900 |
||||||
|
query_dim = 4 |
||||||
|
num_patterns = 0 |
||||||
|
num_feature_levels = 4 |
||||||
|
enc_n_points = 4 |
||||||
|
dec_n_points = 4 |
||||||
|
two_stage_type = "standard" |
||||||
|
two_stage_bbox_embed_share = False |
||||||
|
two_stage_class_embed_share = False |
||||||
|
transformer_activation = "relu" |
||||||
|
dec_pred_bbox_embed_share = True |
||||||
|
dn_box_noise_scale = 1.0 |
||||||
|
dn_label_noise_ratio = 0.5 |
||||||
|
dn_label_coef = 1.0 |
||||||
|
dn_bbox_coef = 1.0 |
||||||
|
embed_init_tgt = True |
||||||
|
dn_labelbook_size = 2000 |
||||||
|
max_text_len = 256 |
||||||
|
text_encoder_type = "bert-base-uncased" |
||||||
|
use_text_enhancer = True |
||||||
|
use_fusion_layer = True |
||||||
|
use_checkpoint = True |
||||||
|
use_transformer_ckpt = True |
||||||
|
use_text_cross_attention = True |
||||||
|
text_dropout = 0.0 |
||||||
|
fusion_dropout = 0.0 |
||||||
|
fusion_droppath = 0.1 |
||||||
|
sub_sentence_present = True |
||||||
@ -0,0 +1,242 @@ |
|||||||
|
from typing import Tuple, List |
||||||
|
|
||||||
|
import cv2 |
||||||
|
import numpy as np |
||||||
|
import supervision as sv |
||||||
|
import torch |
||||||
|
from PIL import Image |
||||||
|
from torchvision.ops import box_convert |
||||||
|
|
||||||
|
import groundingdino.datasets.transforms as T |
||||||
|
from groundingdino.models import build_model |
||||||
|
from groundingdino.util.misc import clean_state_dict |
||||||
|
from groundingdino.util.slconfig import SLConfig |
||||||
|
from groundingdino.util.utils import get_phrases_from_posmap |
||||||
|
|
||||||
|
# ---------------------------------------------------------------------------------------------------------------------- |
||||||
|
# OLD API |
||||||
|
# ---------------------------------------------------------------------------------------------------------------------- |
||||||
|
|
||||||
|
|
||||||
|
def preprocess_caption(caption: str) -> str: |
||||||
|
result = caption.lower().strip() |
||||||
|
if result.endswith("."): |
||||||
|
return result |
||||||
|
return result + "." |
||||||
|
|
||||||
|
|
||||||
|
def load_model(model_config_path: str, model_checkpoint_path: str, device: str = "cuda"): |
||||||
|
args = SLConfig.fromfile(model_config_path) |
||||||
|
args.device = device |
||||||
|
model = build_model(args) |
||||||
|
checkpoint = torch.load(model_checkpoint_path, map_location="cpu") |
||||||
|
model.load_state_dict(clean_state_dict(checkpoint["model"]), strict=False) |
||||||
|
model.eval() |
||||||
|
return model |
||||||
|
|
||||||
|
|
||||||
|
def load_image(image_path: str) -> Tuple[np.array, torch.Tensor]: |
||||||
|
transform = T.Compose( |
||||||
|
[ |
||||||
|
T.RandomResize([800], max_size=1333), |
||||||
|
T.ToTensor(), |
||||||
|
T.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]), |
||||||
|
] |
||||||
|
) |
||||||
|
image_source = Image.open(image_path).convert("RGB") |
||||||
|
image = np.asarray(image_source) |
||||||
|
image_transformed, _ = transform(image_source, None) |
||||||
|
return image, image_transformed |
||||||
|
|
||||||
|
|
||||||
|
def predict( |
||||||
|
model, |
||||||
|
image: torch.Tensor, |
||||||
|
caption: str, |
||||||
|
box_threshold: float, |
||||||
|
text_threshold: float, |
||||||
|
device: str = "cuda" |
||||||
|
) -> Tuple[torch.Tensor, torch.Tensor, List[str]]: |
||||||
|
caption = preprocess_caption(caption=caption) |
||||||
|
|
||||||
|
model = model.to(device) |
||||||
|
image = image.to(device) |
||||||
|
|
||||||
|
with torch.no_grad(): |
||||||
|
outputs = model(image[None], captions=[caption]) |
||||||
|
|
||||||
|
prediction_logits = outputs["pred_logits"].cpu().sigmoid()[0] # prediction_logits.shape = (nq, 256) |
||||||
|
prediction_boxes = outputs["pred_boxes"].cpu()[0] # prediction_boxes.shape = (nq, 4) |
||||||
|
|
||||||
|
mask = prediction_logits.max(dim=1)[0] > box_threshold |
||||||
|
logits = prediction_logits[mask] # logits.shape = (n, 256) |
||||||
|
boxes = prediction_boxes[mask] # boxes.shape = (n, 4) |
||||||
|
|
||||||
|
tokenizer = model.tokenizer |
||||||
|
tokenized = tokenizer(caption) |
||||||
|
|
||||||
|
phrases = [ |
||||||
|
get_phrases_from_posmap(logit > text_threshold, tokenized, tokenizer).replace('.', '') |
||||||
|
for logit |
||||||
|
in logits |
||||||
|
] |
||||||
|
|
||||||
|
return boxes, logits.max(dim=1)[0], phrases |
||||||
|
|
||||||
|
|
||||||
|
def annotate(image_source: np.ndarray, boxes: torch.Tensor, logits: torch.Tensor, phrases: List[str]) -> np.ndarray: |
||||||
|
h, w, _ = image_source.shape |
||||||
|
boxes = boxes * torch.Tensor([w, h, w, h]) |
||||||
|
xyxy = box_convert(boxes=boxes, in_fmt="cxcywh", out_fmt="xyxy").numpy() |
||||||
|
detections = sv.Detections(xyxy=xyxy) |
||||||
|
|
||||||
|
labels = [ |
||||||
|
f"{phrase} {logit:.2f}" |
||||||
|
for phrase, logit |
||||||
|
in zip(phrases, logits) |
||||||
|
] |
||||||
|
|
||||||
|
box_annotator = sv.BoxAnnotator() |
||||||
|
annotated_frame = cv2.cvtColor(image_source, cv2.COLOR_RGB2BGR) |
||||||
|
annotated_frame = box_annotator.annotate(scene=annotated_frame, detections=detections, labels=labels) |
||||||
|
return annotated_frame |
||||||
|
|
||||||
|
|
||||||
|
# ---------------------------------------------------------------------------------------------------------------------- |
||||||
|
# NEW API |
||||||
|
# ---------------------------------------------------------------------------------------------------------------------- |
||||||
|
|
||||||
|
|
||||||
|
class Model: |
||||||
|
|
||||||
|
def __init__( |
||||||
|
self, |
||||||
|
model_config_path: str, |
||||||
|
model_checkpoint_path: str, |
||||||
|
device: str = "cuda" |
||||||
|
): |
||||||
|
self.model = load_model( |
||||||
|
model_config_path=model_config_path, |
||||||
|
model_checkpoint_path=model_checkpoint_path, |
||||||
|
device=device |
||||||
|
).to(device) |
||||||
|
self.device = device |
||||||
|
|
||||||
|
def predict_with_caption( |
||||||
|
self, |
||||||
|
image: np.ndarray, |
||||||
|
caption: str, |
||||||
|
box_threshold: float = 0.35, |
||||||
|
text_threshold: float = 0.25 |
||||||
|
) -> Tuple[sv.Detections, List[str]]: |
||||||
|
""" |
||||||
|
import cv2 |
||||||
|
|
||||||
|
image = cv2.imread(IMAGE_PATH) |
||||||
|
|
||||||
|
model = Model(model_config_path=CONFIG_PATH, model_checkpoint_path=WEIGHTS_PATH) |
||||||
|
detections, labels = model.predict_with_caption( |
||||||
|
image=image, |
||||||
|
caption=caption, |
||||||
|
box_threshold=BOX_THRESHOLD, |
||||||
|
text_threshold=TEXT_THRESHOLD |
||||||
|
) |
||||||
|
|
||||||
|
import supervision as sv |
||||||
|
|
||||||
|
box_annotator = sv.BoxAnnotator() |
||||||
|
annotated_image = box_annotator.annotate(scene=image, detections=detections, labels=labels) |
||||||
|
""" |
||||||
|
processed_image = Model.preprocess_image(image_bgr=image).to(self.device) |
||||||
|
boxes, logits, phrases = predict( |
||||||
|
model=self.model, |
||||||
|
image=processed_image, |
||||||
|
caption=caption, |
||||||
|
box_threshold=box_threshold, |
||||||
|
text_threshold=text_threshold) |
||||||
|
source_h, source_w, _ = image.shape |
||||||
|
detections = Model.post_process_result( |
||||||
|
source_h=source_h, |
||||||
|
source_w=source_w, |
||||||
|
boxes=boxes, |
||||||
|
logits=logits) |
||||||
|
return detections, phrases |
||||||
|
|
||||||
|
def predict_with_classes( |
||||||
|
self, |
||||||
|
image: np.ndarray, |
||||||
|
classes: List[str], |
||||||
|
box_threshold: float, |
||||||
|
text_threshold: float |
||||||
|
) -> sv.Detections: |
||||||
|
""" |
||||||
|
import cv2 |
||||||
|
|
||||||
|
image = cv2.imread(IMAGE_PATH) |
||||||
|
|
||||||
|
model = Model(model_config_path=CONFIG_PATH, model_checkpoint_path=WEIGHTS_PATH) |
||||||
|
detections = model.predict_with_classes( |
||||||
|
image=image, |
||||||
|
classes=CLASSES, |
||||||
|
box_threshold=BOX_THRESHOLD, |
||||||
|
text_threshold=TEXT_THRESHOLD |
||||||
|
) |
||||||
|
|
||||||
|
|
||||||
|
import supervision as sv |
||||||
|
|
||||||
|
box_annotator = sv.BoxAnnotator() |
||||||
|
annotated_image = box_annotator.annotate(scene=image, detections=detections) |
||||||
|
""" |
||||||
|
caption = ", ".join(classes) |
||||||
|
processed_image = Model.preprocess_image(image_bgr=image).to(self.device) |
||||||
|
boxes, logits, phrases = predict( |
||||||
|
model=self.model, |
||||||
|
image=processed_image, |
||||||
|
caption=caption, |
||||||
|
box_threshold=box_threshold, |
||||||
|
text_threshold=text_threshold) |
||||||
|
source_h, source_w, _ = image.shape |
||||||
|
detections = Model.post_process_result( |
||||||
|
source_h=source_h, |
||||||
|
source_w=source_w, |
||||||
|
boxes=boxes, |
||||||
|
logits=logits) |
||||||
|
class_id = Model.phrases2classes(phrases=phrases, classes=classes) |
||||||
|
detections.class_id = class_id |
||||||
|
return detections |
||||||
|
|
||||||
|
@staticmethod |
||||||
|
def preprocess_image(image_bgr: np.ndarray) -> torch.Tensor: |
||||||
|
transform = T.Compose( |
||||||
|
[ |
||||||
|
T.RandomResize([800], max_size=1333), |
||||||
|
T.ToTensor(), |
||||||
|
T.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]), |
||||||
|
] |
||||||
|
) |
||||||
|
image_pillow = Image.fromarray(cv2.cvtColor(image_bgr, cv2.COLOR_BGR2RGB)) |
||||||
|
image_transformed, _ = transform(image_pillow, None) |
||||||
|
return image_transformed |
||||||
|
|
||||||
|
@staticmethod |
||||||
|
def post_process_result( |
||||||
|
source_h: int, |
||||||
|
source_w: int, |
||||||
|
boxes: torch.Tensor, |
||||||
|
logits: torch.Tensor |
||||||
|
) -> sv.Detections: |
||||||
|
boxes = boxes * torch.Tensor([source_w, source_h, source_w, source_h]) |
||||||
|
xyxy = box_convert(boxes=boxes, in_fmt="cxcywh", out_fmt="xyxy").numpy() |
||||||
|
confidence = logits.numpy() |
||||||
|
return sv.Detections(xyxy=xyxy, confidence=confidence) |
||||||
|
|
||||||
|
@staticmethod |
||||||
|
def phrases2classes(phrases: List[str], classes: List[str]) -> np.ndarray: |
||||||
|
class_ids = [] |
||||||
|
for phrase in phrases: |
||||||
|
try: |
||||||
|
class_ids.append(classes.index(phrase)) |
||||||
|
except ValueError: |
||||||
|
class_ids.append(None) |
||||||
|
return np.array(class_ids) |
||||||
@ -1 +1,10 @@ |
|||||||
transformers==4.5.1 |
torch |
||||||
|
torchvision |
||||||
|
transformers |
||||||
|
addict |
||||||
|
yapf |
||||||
|
timm |
||||||
|
numpy |
||||||
|
opencv-python |
||||||
|
supervision==0.4.0 |
||||||
|
pycocotools |
||||||
Loading…
Reference in new issue