Testing if maskformer is working
Browse files
app.py
CHANGED
@@ -5,115 +5,66 @@ import requests, validators
|
|
5 |
import torch
|
6 |
import pathlib
|
7 |
from PIL import Image
|
8 |
-
|
9 |
-
from
|
|
|
10 |
|
11 |
import os
|
12 |
|
13 |
-
# colors for visualization
|
14 |
-
COLORS = [
|
15 |
-
[0.000, 0.447, 0.741],
|
16 |
-
[0.850, 0.325, 0.098],
|
17 |
-
[0.929, 0.694, 0.125],
|
18 |
-
[0.494, 0.184, 0.556],
|
19 |
-
[0.466, 0.674, 0.188],
|
20 |
-
[0.301, 0.745, 0.933]
|
21 |
-
]
|
22 |
-
|
23 |
-
YOLOV8_LABELS = ['pedestrian', 'people', 'bicycle', 'car', 'van', 'truck', 'tricycle', 'awning-tricycle', 'bus', 'motor']
|
24 |
-
|
25 |
-
def make_prediction(img, feature_extractor, model):
|
26 |
-
inputs = feature_extractor(img, return_tensors="pt")
|
27 |
-
outputs = model(**inputs)
|
28 |
-
img_size = torch.tensor([tuple(reversed(img.size))])
|
29 |
-
processed_outputs = feature_extractor.post_process(outputs, img_size)
|
30 |
-
return processed_outputs
|
31 |
-
|
32 |
-
def fig2img(fig):
|
33 |
-
buf = io.BytesIO()
|
34 |
-
fig.savefig(buf, bbox_inches="tight")
|
35 |
-
buf.seek(0)
|
36 |
-
img = Image.open(buf)
|
37 |
-
return img
|
38 |
-
|
39 |
-
|
40 |
-
def visualize_prediction(pil_img, output_dict, threshold=0.7, id2label=None):
|
41 |
-
keep = output_dict["scores"] > threshold
|
42 |
-
boxes = output_dict["boxes"][keep].tolist()
|
43 |
-
scores = output_dict["scores"][keep].tolist()
|
44 |
-
labels = output_dict["labels"][keep].tolist()
|
45 |
-
if id2label is not None:
|
46 |
-
labels = [id2label[x] for x in labels]
|
47 |
-
|
48 |
-
# print("Labels " + str(labels))
|
49 |
-
|
50 |
-
plt.figure(figsize=(16, 10))
|
51 |
-
plt.imshow(pil_img)
|
52 |
-
ax = plt.gca()
|
53 |
-
colors = COLORS * 100
|
54 |
-
for score, (xmin, ymin, xmax, ymax), label, color in zip(scores, boxes, labels, colors):
|
55 |
-
ax.add_patch(plt.Rectangle((xmin, ymin), xmax - xmin, ymax - ymin, fill=False, color=color, linewidth=3))
|
56 |
-
ax.text(xmin, ymin, f"{label}: {score:0.2f}", fontsize=15, bbox=dict(facecolor="yellow", alpha=0.5))
|
57 |
-
plt.axis("off")
|
58 |
-
return fig2img(plt.gcf())
|
59 |
|
60 |
def detect_objects(model_name,url_input,image_input,threshold):
|
61 |
|
62 |
|
63 |
-
if '
|
64 |
-
|
65 |
-
|
66 |
-
|
67 |
-
|
68 |
-
|
69 |
-
|
70 |
-
|
71 |
-
model.overrides['agnostic_nms'] = False # NMS class-agnostic
|
72 |
-
model.overrides['max_det'] = 1000 # maximum number of detections per image
|
73 |
-
|
74 |
-
results = model.predict(image_input)
|
75 |
-
|
76 |
-
render = render_result(model=model, image=image_input, result=results[0])
|
77 |
-
|
78 |
-
final_str = ""
|
79 |
-
final_str_abv = ""
|
80 |
-
final_str_else = ""
|
81 |
-
|
82 |
-
for result in results:
|
83 |
-
boxes = result.boxes.cpu().numpy()
|
84 |
-
for i, box in enumerate(boxes):
|
85 |
-
# r = box.xyxy[0].astype(int)
|
86 |
-
coordinates = box.xyxy[0].astype(int)
|
87 |
-
try:
|
88 |
-
label = YOLOV8_LABELS[int(box.cls)]
|
89 |
-
except:
|
90 |
-
label = "ERROR"
|
91 |
-
try:
|
92 |
-
confi = float(box.conf)
|
93 |
-
except:
|
94 |
-
confi = 0.0
|
95 |
-
# final_str_abv += str() + "__" + str(box.cls) + "__" + str(box.conf) + "__" + str(box) + "\n"
|
96 |
-
if confi >= threshold:
|
97 |
-
final_str_abv += f"Detected `{label}` with confidence `{confi}` at location `{coordinates}`\n"
|
98 |
-
else:
|
99 |
-
final_str_else += f"Detected `{label}` with confidence `{confi}` at location `{coordinates}`\n"
|
100 |
-
|
101 |
-
final_str = "{:*^50}\n".format("ABOVE THRESHOLD OR EQUAL") + final_str_abv + "\n{:*^50}\n".format("BELOW THRESHOLD")+final_str_else
|
102 |
-
|
103 |
-
return render, final_str
|
104 |
-
else:
|
105 |
-
|
106 |
-
#Extract model and feature extractor
|
107 |
-
feature_extractor = AutoFeatureExtractor.from_pretrained(model_name)
|
108 |
-
if 'detr' in model_name:
|
109 |
-
|
110 |
-
model = DetrForObjectDetection.from_pretrained(model_name)
|
111 |
|
112 |
-
|
113 |
-
|
114 |
-
|
115 |
-
|
116 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
117 |
if validators.url(url_input):
|
118 |
image = Image.open(requests.get(url_input, stream=True).raw)
|
119 |
tb_label = "Confidence Values URL"
|
@@ -121,11 +72,28 @@ def detect_objects(model_name,url_input,image_input,threshold):
|
|
121 |
elif image_input:
|
122 |
image = image_input
|
123 |
tb_label = "Confidence Values Upload"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
124 |
|
125 |
-
|
126 |
-
|
127 |
-
|
128 |
-
|
129 |
|
130 |
#Visualize prediction
|
131 |
viz_img = visualize_prediction(image, processed_outputs, threshold, model.config.id2label)
|
@@ -146,6 +114,8 @@ def detect_objects(model_name,url_input,image_input,threshold):
|
|
146 |
final_str = "{:*^50}\n".format("ABOVE THRESHOLD OR EQUAL") + final_str_abv + "\n{:*^50}\n".format("BELOW THRESHOLD")+final_str_else
|
147 |
|
148 |
return viz_img, final_str
|
|
|
|
|
149 |
|
150 |
def set_example_image(example: list) -> dict:
|
151 |
return gr.Image.update(value=example[0])
|
@@ -162,16 +132,9 @@ Links to HuggingFace Models:
|
|
162 |
- [facebook/detr-resnet-50-panoptic](https://huggingface.co/facebook/detr-resnet-50-panoptic)
|
163 |
- [facebook/detr-resnet-101-panoptic](https://huggingface.co/facebook/detr-resnet-101-panoptic)
|
164 |
- [facebook/maskformer-swin-large-coco](https://huggingface.co/facebook/maskformer-swin-large-coco)
|
165 |
-
- [hustvl/yolos-small](https://huggingface.co/hustvl/yolos-small)
|
166 |
-
- [hustvl/yolos-tiny](https://huggingface.co/hustvl/yolos-tiny)
|
167 |
-
- [facebook/detr-resnet-101-dc5](https://huggingface.co/facebook/detr-resnet-101-dc5)
|
168 |
-
- [hustvl/yolos-small-300](https://huggingface.co/hustvl/yolos-small-300)
|
169 |
-
- [mshamrai/yolov8x-visdrone](https://huggingface.co/mshamrai/yolov8x-visdrone)
|
170 |
-
|
171 |
"""
|
172 |
|
173 |
-
models = ["facebook/detr-resnet-50-panoptic","facebook/detr-resnet-101-panoptic","facebook/maskformer-swin-large-coco"
|
174 |
-
'hustvl/yolos-small','hustvl/yolos-tiny','facebook/detr-resnet-101-dc5', 'hustvl/yolos-small-300', 'mshamrai/yolov8x-visdrone']
|
175 |
urls = ["https://c8.alamy.com/comp/J2AB4K/the-new-york-stock-exchange-on-the-wall-street-in-new-york-J2AB4K.jpg"]
|
176 |
|
177 |
# twitter_link = """
|
@@ -196,7 +159,7 @@ with demo:
|
|
196 |
gr.Markdown(title)
|
197 |
gr.Markdown(description)
|
198 |
# gr.Markdown(twitter_link)
|
199 |
-
options = gr.Dropdown(choices=models,label='Select
|
200 |
|
201 |
slider_input = gr.Slider(minimum=0.2,maximum=1,value=0.7,label='Prediction Threshold')
|
202 |
|
|
|
5 |
import torch
|
6 |
import pathlib
|
7 |
from PIL import Image
|
8 |
+
|
9 |
+
from transformers import DetrFeatureExtractor, DetrForSegmentation
|
10 |
+
from transformers.models.detr.feature_extraction_detr import rgb_to_id
|
11 |
|
12 |
import os
|
13 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
14 |
|
15 |
def detect_objects(model_name,url_input,image_input,threshold):
|
16 |
|
17 |
|
18 |
+
if 'maskformer' in model_name:
|
19 |
+
if validators.url(url_input):
|
20 |
+
image = Image.open(requests.get(url_input, stream=True).raw)
|
21 |
+
tb_label = "Confidence Values URL"
|
22 |
+
|
23 |
+
elif image_input:
|
24 |
+
image = image_input
|
25 |
+
tb_label = "Confidence Values Upload"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
26 |
|
27 |
+
# NOTE: Pulling from the example on https://huggingface.co/facebook/maskformer-swin-large-coco
|
28 |
+
# and https://huggingface.co/spaces/ajcdp/Image-Segmentation-Gradio/blob/main/app.py
|
29 |
+
|
30 |
+
processor = MaskFormerImageProcessor.from_pretrained(model_name)
|
31 |
+
model = MaskFormerForInstanceSegmentation.from_pretrained(model_name)
|
32 |
+
|
33 |
+
target_size = (img.shape[0], img.shape[1])
|
34 |
+
inputs = preprocessor(images=img, return_tensors="pt")
|
35 |
+
with torch.no_grad():
|
36 |
+
outputs = model(**inputs)
|
37 |
+
outputs.class_queries_logits = outputs.class_queries_logits.cpu()
|
38 |
+
outputs.masks_queries_logits = outputs.masks_queries_logits.cpu()
|
39 |
+
results = preprocessor.post_process_segmentation(outputs=outputs, target_size=target_size)[0].cpu().detach()
|
40 |
+
results = torch.argmax(results, dim=0).numpy()
|
41 |
+
results = visualize_instance_seg_mask(results)
|
42 |
+
return results, "EMPTY"
|
43 |
+
|
44 |
+
# for result in results:
|
45 |
+
# boxes = result.boxes.cpu().numpy()
|
46 |
+
# for i, box in enumerate(boxes):
|
47 |
+
# # r = box.xyxy[0].astype(int)
|
48 |
+
# coordinates = box.xyxy[0].astype(int)
|
49 |
+
# try:
|
50 |
+
# label = YOLOV8_LABELS[int(box.cls)]
|
51 |
+
# except:
|
52 |
+
# label = "ERROR"
|
53 |
+
# try:
|
54 |
+
# confi = float(box.conf)
|
55 |
+
# except:
|
56 |
+
# confi = 0.0
|
57 |
+
# # final_str_abv += str() + "__" + str(box.cls) + "__" + str(box.conf) + "__" + str(box) + "\n"
|
58 |
+
# if confi >= threshold:
|
59 |
+
# final_str_abv += f"Detected `{label}` with confidence `{confi}` at location `{coordinates}`\n"
|
60 |
+
# else:
|
61 |
+
# final_str_else += f"Detected `{label}` with confidence `{confi}` at location `{coordinates}`\n"
|
62 |
+
|
63 |
+
# final_str = "{:*^50}\n".format("ABOVE THRESHOLD OR EQUAL") + final_str_abv + "\n{:*^50}\n".format("BELOW THRESHOLD")+final_str_else
|
64 |
+
|
65 |
+
# return render, final_str
|
66 |
+
elif "detr" in model_name:
|
67 |
+
# NOTE: Using the example on https://huggingface.co/facebook/detr-resnet-50-panoptic
|
68 |
if validators.url(url_input):
|
69 |
image = Image.open(requests.get(url_input, stream=True).raw)
|
70 |
tb_label = "Confidence Values URL"
|
|
|
72 |
elif image_input:
|
73 |
image = image_input
|
74 |
tb_label = "Confidence Values Upload"
|
75 |
+
|
76 |
+
feature_extractor = DetrFeatureExtractor.from_pretrained(model_name)
|
77 |
+
model = DetrForSegmentation.from_pretrained(model_name)
|
78 |
+
inputs = feature_extractor(images=image, return_tensors="pt")
|
79 |
+
|
80 |
+
outputs = model(**inputs)
|
81 |
+
|
82 |
+
# use the `post_process_panoptic` method of `DetrFeatureExtractor` to convert to COCO format
|
83 |
+
processed_sizes = torch.as_tensor(inputs["pixel_values"].shape[-2:]).unsqueeze(0)
|
84 |
+
result = feature_extractor.post_process_panoptic(outputs, processed_sizes)[0]
|
85 |
+
|
86 |
+
# the segmentation is stored in a special-format png
|
87 |
+
panoptic_seg = Image.open(io.BytesIO(result["png_string"]))
|
88 |
+
panoptic_seg = numpy.array(panoptic_seg, dtype=numpy.uint8)
|
89 |
+
# retrieve the ids corresponding to each mask
|
90 |
+
panoptic_seg_id = rgb_to_id(panoptic_seg)
|
91 |
+
|
92 |
|
93 |
+
|
94 |
+
return gr.Image.update(), "EMPTY"
|
95 |
+
|
96 |
+
|
97 |
|
98 |
#Visualize prediction
|
99 |
viz_img = visualize_prediction(image, processed_outputs, threshold, model.config.id2label)
|
|
|
114 |
final_str = "{:*^50}\n".format("ABOVE THRESHOLD OR EQUAL") + final_str_abv + "\n{:*^50}\n".format("BELOW THRESHOLD")+final_str_else
|
115 |
|
116 |
return viz_img, final_str
|
117 |
+
else:
|
118 |
+
raise NameError(f"Model name {model_name} not prepared")
|
119 |
|
120 |
def set_example_image(example: list) -> dict:
|
121 |
return gr.Image.update(value=example[0])
|
|
|
132 |
- [facebook/detr-resnet-50-panoptic](https://huggingface.co/facebook/detr-resnet-50-panoptic)
|
133 |
- [facebook/detr-resnet-101-panoptic](https://huggingface.co/facebook/detr-resnet-101-panoptic)
|
134 |
- [facebook/maskformer-swin-large-coco](https://huggingface.co/facebook/maskformer-swin-large-coco)
|
|
|
|
|
|
|
|
|
|
|
|
|
135 |
"""
|
136 |
|
137 |
+
models = ["facebook/detr-resnet-50-panoptic","facebook/detr-resnet-101-panoptic","facebook/maskformer-swin-large-coco"]
|
|
|
138 |
urls = ["https://c8.alamy.com/comp/J2AB4K/the-new-york-stock-exchange-on-the-wall-street-in-new-york-J2AB4K.jpg"]
|
139 |
|
140 |
# twitter_link = """
|
|
|
159 |
gr.Markdown(title)
|
160 |
gr.Markdown(description)
|
161 |
# gr.Markdown(twitter_link)
|
162 |
+
options = gr.Dropdown(choices=models,label='Select Image Segmentation Model',show_label=True)
|
163 |
|
164 |
slider_input = gr.Slider(minimum=0.2,maximum=1,value=0.7,label='Prediction Threshold')
|
165 |
|