Mar2Ding commited on
Commit
d6ee06c
1 Parent(s): 1115e1b

Update setup.py

Browse files
Files changed (1) hide show
  1. setup.py +68 -632
setup.py CHANGED
@@ -1,636 +1,72 @@
1
- import subprocess
2
- import re
3
- from typing import List, Tuple, Optional
4
- import os
5
-
6
-
7
- # Define the command to be executed
8
- command = ["python", "setup.py", "build_ext", "--inplace"]
9
-
10
- # Execute the command
11
- result = subprocess.run(command, capture_output=True, text=True)
12
-
13
-
14
-
15
-
16
- def install_cuda_toolkit():
17
- # CUDA_TOOLKIT_URL = "https://developer.download.nvidia.com/compute/cuda/11.8.0/local_installers/cuda_11.8.0_520.61.05_linux.run"
18
- CUDA_TOOLKIT_URL = "https://developer.download.nvidia.com/compute/cuda/12.2.0/local_installers/cuda_12.2.0_535.54.03_linux.run"
19
- CUDA_TOOLKIT_FILE = "/tmp/%s" % os.path.basename(CUDA_TOOLKIT_URL)
20
- subprocess.call(["wget", "-q", CUDA_TOOLKIT_URL, "-O", CUDA_TOOLKIT_FILE])
21
- subprocess.call(["chmod", "+x", CUDA_TOOLKIT_FILE])
22
- subprocess.call([CUDA_TOOLKIT_FILE, "--silent", "--toolkit"])
23
-
24
- os.environ["CUDA_HOME"] = "/usr/local/cuda"
25
- os.environ["PATH"] = "%s/bin:%s" % (os.environ["CUDA_HOME"], os.environ["PATH"])
26
- os.environ["LD_LIBRARY_PATH"] = "%s/lib:%s" % (
27
- os.environ["CUDA_HOME"],
28
- "" if "LD_LIBRARY_PATH" not in os.environ else os.environ["LD_LIBRARY_PATH"],
29
- )
30
- # Fix: arch_list[-1] += '+PTX'; IndexError: list index out of range
31
- os.environ["TORCH_CUDA_ARCH_LIST"] = "8.0;8.6"
32
-
33
- install_cuda_toolkit()
34
-
35
- css="""
36
- div#component-18, div#component-25, div#component-35, div#component-41{
37
- align-items: stretch!important;
38
  }
39
- """
40
-
41
- # Print the output and error (if any)
42
- print("Output:\n", result.stdout)
43
- print("Errors:\n", result.stderr)
44
-
45
- # Check if the command was successful
46
- if result.returncode == 0:
47
- print("Command executed successfully.")
48
- else:
49
- print("Command failed with return code:", result.returncode)
50
-
51
- import gradio as gr
52
- from datetime import datetime
53
- os.environ["TORCH_CUDNN_SDPA_ENABLED"] = "1"
54
- import torch
55
- import numpy as np
56
- import cv2
57
- import matplotlib.pyplot as plt
58
- from PIL import Image, ImageFilter
59
- from sam2.build_sam import build_sam2_video_predictor
60
-
61
- from moviepy.editor import ImageSequenceClip
62
-
63
- def get_video_fps(video_path):
64
- # Open the video file
65
- cap = cv2.VideoCapture(video_path)
66
-
67
- if not cap.isOpened():
68
- print("Error: Could not open video.")
69
- return None
70
-
71
- # Get the FPS of the video
72
- fps = cap.get(cv2.CAP_PROP_FPS)
73
-
74
- return fps
75
-
76
- def clear_points(image):
77
- # we clean all
78
- return [
79
- image, # first_frame_path
80
- gr.State([]), # tracking_points
81
- gr.State([]), # trackings_input_label
82
- image, # points_map
83
- #gr.State() # stored_inference_state
84
- ]
85
-
86
- def preprocess_video_in(video_path):
87
-
88
- # Generate a unique ID based on the current date and time
89
- unique_id = datetime.now().strftime('%Y%m%d%H%M%S')
90
-
91
- # Set directory with this ID to store video frames
92
- extracted_frames_output_dir = f'frames_{unique_id}'
93
-
94
- # Create the output directory
95
- os.makedirs(extracted_frames_output_dir, exist_ok=True)
96
-
97
- ### Process video frames ###
98
- # Open the video file
99
- cap = cv2.VideoCapture(video_path)
100
-
101
- if not cap.isOpened():
102
- print("Error: Could not open video.")
103
- return None
104
-
105
- # Get the frames per second (FPS) of the video
106
- fps = cap.get(cv2.CAP_PROP_FPS)
107
-
108
- # Calculate the number of frames to process (10 seconds of video)
109
- max_frames = int(fps * 10)
110
-
111
- frame_number = 0
112
- first_frame = None
113
-
114
- while True:
115
- ret, frame = cap.read()
116
- if not ret or frame_number >= max_frames:
117
- break
118
-
119
- # Format the frame filename as '00000.jpg'
120
- frame_filename = os.path.join(extracted_frames_output_dir, f'{frame_number:05d}.jpg')
121
-
122
- # Save the frame as a JPEG file
123
- cv2.imwrite(frame_filename, frame)
124
-
125
- # Store the first frame
126
- if frame_number == 0:
127
- first_frame = frame_filename
128
-
129
- frame_number += 1
130
-
131
- # Release the video capture object
132
- cap.release()
133
-
134
- # scan all the JPEG frame names in this directory
135
- scanned_frames = [
136
- p for p in os.listdir(extracted_frames_output_dir)
137
- if os.path.splitext(p)[-1] in [".jpg", ".jpeg", ".JPG", ".JPEG"]
138
- ]
139
- scanned_frames.sort(key=lambda p: int(os.path.splitext(p)[0]))
140
- # print(f"SCANNED_FRAMES: {scanned_frames}")
141
-
142
- return [
143
- first_frame, # first_frame_path
144
- gr.State([]), # tracking_points
145
- gr.State([]), # trackings_input_label
146
- first_frame, # input_first_frame_image
147
- first_frame, # points_map
148
- extracted_frames_output_dir, # video_frames_dir
149
- scanned_frames, # scanned_frames
150
- None, # stored_inference_state
151
- None, # stored_frame_names
152
- gr.update(open=False) # video_in_drawer
153
- ]
154
-
155
- def get_point(point_type, tracking_points, trackings_input_label, input_first_frame_image, evt: gr.SelectData):
156
- print(f"You selected {evt.value} at {evt.index} from {evt.target}")
157
-
158
- tracking_points.value.append(evt.index)
159
- print(f"TRACKING POINT: {tracking_points.value}")
160
-
161
- if point_type == "include":
162
- trackings_input_label.value.append(1)
163
- elif point_type == "exclude":
164
- trackings_input_label.value.append(0)
165
- print(f"TRACKING INPUT LABEL: {trackings_input_label.value}")
166
-
167
- # Open the image and get its dimensions
168
- transparent_background = Image.open(input_first_frame_image).convert('RGBA')
169
- w, h = transparent_background.size
170
-
171
- # Define the circle radius as a fraction of the smaller dimension
172
- fraction = 0.02 # You can adjust this value as needed
173
- radius = int(fraction * min(w, h))
174
-
175
- # Create a transparent layer to draw on
176
- transparent_layer = np.zeros((h, w, 4), dtype=np.uint8)
177
-
178
- for index, track in enumerate(tracking_points.value):
179
- if trackings_input_label.value[index] == 1:
180
- cv2.circle(transparent_layer, track, radius, (0, 255, 0, 255), -1)
181
- else:
182
- cv2.circle(transparent_layer, track, radius, (255, 0, 0, 255), -1)
183
-
184
- # Convert the transparent layer back to an image
185
- transparent_layer = Image.fromarray(transparent_layer, 'RGBA')
186
- selected_point_map = Image.alpha_composite(transparent_background, transparent_layer)
187
-
188
- return tracking_points, trackings_input_label, selected_point_map
189
-
190
- # use bfloat16 for the entire notebook
191
- torch.autocast(device_type="cuda", dtype=torch.bfloat16).__enter__()
192
-
193
- if torch.cuda.get_device_properties(0).major >= 8:
194
- # turn on tfloat32 for Ampere GPUs (https://pytorch.org/docs/stable/notes/cuda.html#tensorfloat-32-tf32-on-ampere-devices)
195
- torch.backends.cuda.matmul.allow_tf32 = True
196
- torch.backends.cudnn.allow_tf32 = True
197
-
198
- def show_mask(mask, ax, obj_id=None, random_color=False):
199
- if random_color:
200
- color = np.concatenate([np.random.random(3), np.array([0.6])], axis=0)
201
- else:
202
- cmap = plt.get_cmap("tab10")
203
- cmap_idx = 0 if obj_id is None else obj_id
204
- color = np.array([*cmap(cmap_idx)[:3], 0.6])
205
- h, w = mask.shape[-2:]
206
- mask_image = mask.reshape(h, w, 1) * color.reshape(1, 1, -1)
207
- ax.imshow(mask_image)
208
-
209
-
210
- def show_points(coords, labels, ax, marker_size=200):
211
- pos_points = coords[labels==1]
212
- neg_points = coords[labels==0]
213
- ax.scatter(pos_points[:, 0], pos_points[:, 1], color='green', marker='*', s=marker_size, edgecolor='white', linewidth=1.25)
214
- ax.scatter(neg_points[:, 0], neg_points[:, 1], color='red', marker='*', s=marker_size, edgecolor='white', linewidth=1.25)
215
-
216
- def show_box(box, ax):
217
- x0, y0 = box[0], box[1]
218
- w, h = box[2] - box[0], box[3] - box[1]
219
- ax.add_patch(plt.Rectangle((x0, y0), w, h, edgecolor='green', facecolor=(0, 0, 0, 0), lw=2))
220
-
221
-
222
- def load_model(checkpoint):
223
- # Load model accordingly to user's choice
224
- if checkpoint == "tiny":
225
- sam2_checkpoint = "./checkpoints/sam2.1_hiera_tiny.pt"
226
- model_cfg = "configs/sam2.1/sam2.1_hiera_t.yaml"
227
- return [sam2_checkpoint, model_cfg]
228
- elif checkpoint == "samll":
229
- sam2_checkpoint = "./checkpoints/sam2.1_hiera_small.pt"
230
- model_cfg = "configs/sam2.1/sam2.1_hiera_s.yaml"
231
- return [sam2_checkpoint, model_cfg]
232
- elif checkpoint == "base-plus":
233
- sam2_checkpoint = "./checkpoints/sam2.1_hiera_base_plus.pt"
234
- model_cfg = "configs/sam2.1/sam2.1_hiera_b+.yaml"
235
- return [sam2_checkpoint, model_cfg]
236
- # elif checkpoint == "large":
237
- # sam2_checkpoint = "./checkpoints/sam2.1_hiera_large.pt"
238
- # model_cfg = "configs/sam2.1/sam2.1_hiera_l.yaml"
239
- # return [sam2_checkpoint, model_cfg]
240
-
241
-
242
-
243
- def get_mask_sam_process(
244
- stored_inference_state,
245
- input_first_frame_image,
246
- checkpoint,
247
- tracking_points,
248
- trackings_input_label,
249
- video_frames_dir, # extracted_frames_output_dir defined in 'preprocess_video_in' function
250
- scanned_frames,
251
- working_frame: str = None, # current frame being added points
252
- available_frames_to_check: List[str] = [],
253
- # progress=gr.Progress(track_tqdm=True)
254
- ):
255
-
256
- # get model and model config paths
257
- print(f"USER CHOSEN CHECKPOINT: {checkpoint}")
258
- sam2_checkpoint, model_cfg = load_model(checkpoint)
259
- print("MODEL LOADED")
260
-
261
- # set predictor
262
- predictor = build_sam2_video_predictor(model_cfg, sam2_checkpoint)
263
- print("PREDICTOR READY")
264
 
265
- # `video_dir` a directory of JPEG frames with filenames like `<frame_index>.jpg`
266
- # print(f"STATE FRAME OUTPUT DIRECTORY: {video_frames_dir}")
267
- video_dir = video_frames_dir
268
-
269
- # scan all the JPEG frame names in this directory
270
- frame_names = scanned_frames
271
 
272
- # print(f"STORED INFERENCE STEP: {stored_inference_state}")
273
- if stored_inference_state is None:
274
- # Init SAM2 inference_state
275
- inference_state = predictor.init_state(video_path=video_dir)
276
- inference_state['num_pathway'] = 3
277
- inference_state['iou_thre'] = 0.3
278
- inference_state['uncertainty'] = 2
279
- print("NEW INFERENCE_STATE INITIATED")
280
- else:
281
- inference_state = stored_inference_state
282
-
283
- # segment and track one object
284
- # predictor.reset_state(inference_state) # if any previous tracking, reset
285
-
286
-
287
- ### HANDLING WORKING FRAME
288
- # new_working_frame = None
289
- # Add new point
290
- if working_frame is None:
291
- ann_frame_idx = 0 # the frame index we interact with, 0 if it is the first frame
292
- working_frame = "00000.jpg"
293
- else:
294
- # Use a regular expression to find the integer
295
- match = re.search(r'frame_(\d+)', working_frame)
296
- if match:
297
- # Extract the integer from the match
298
- frame_number = int(match.group(1))
299
- ann_frame_idx = frame_number
300
-
301
- print(f"NEW_WORKING_FRAME PATH: {working_frame}")
302
-
303
- ann_obj_id = 1 # give a unique id to each object we interact with (it can be any integers)
304
-
305
- # Let's add a positive click at (x, y) = (210, 350) to get started
306
- points = np.array(tracking_points.value, dtype=np.float32)
307
- # for labels, `1` means positive click and `0` means negative click
308
- labels = np.array(trackings_input_label.value, np.int32)
309
- _, out_obj_ids, out_mask_logits = predictor.add_new_points(
310
- inference_state=inference_state,
311
- frame_idx=ann_frame_idx,
312
- obj_id=ann_obj_id,
313
- points=points,
314
- labels=labels,
315
- )
316
-
317
- # Create the plot
318
- plt.figure(figsize=(12, 8))
319
- plt.title(f"frame {ann_frame_idx}")
320
- plt.imshow(Image.open(os.path.join(video_dir, frame_names[ann_frame_idx])))
321
- show_points(points, labels, plt.gca())
322
- show_mask((out_mask_logits[0] > 0.0).cpu().numpy(), plt.gca(), obj_id=out_obj_ids[0])
323
-
324
- # Save the plot as a JPG file
325
- first_frame_output_filename = "output_first_frame.jpg"
326
- plt.savefig(first_frame_output_filename, format='jpg')
327
- plt.close()
328
- torch.cuda.empty_cache()
329
-
330
- # Assuming available_frames_to_check.value is a list
331
- if working_frame not in available_frames_to_check:
332
- available_frames_to_check.append(working_frame)
333
- print(available_frames_to_check)
334
-
335
- # return gr.update(visible=True), "output_first_frame.jpg", frame_names, predictor, inference_state, gr.update(choices=available_frames_to_check, value=working_frame, visible=True)
336
- return "output_first_frame.jpg", frame_names, predictor, inference_state, gr.update(choices=available_frames_to_check, value=working_frame, visible=False)
337
-
338
- def propagate_to_all(video_in, checkpoint, stored_inference_state, stored_frame_names, video_frames_dir, vis_frame_type, available_frames_to_check, working_frame, progress=gr.Progress(track_tqdm=True)):
339
- #### PROPAGATION ####
340
- sam2_checkpoint, model_cfg = load_model(checkpoint)
341
- predictor = build_sam2_video_predictor(model_cfg, sam2_checkpoint)
342
-
343
- inference_state = stored_inference_state
344
- frame_names = stored_frame_names
345
- video_dir = video_frames_dir
346
-
347
- # Define a directory to save the JPEG images
348
- frames_output_dir = "frames_output_images"
349
- os.makedirs(frames_output_dir, exist_ok=True)
350
-
351
- # Initialize a list to store file paths of saved images
352
- jpeg_images = []
353
-
354
- # run propagation throughout the video and collect the results in a dict
355
- video_segments = {} # video_segments contains the per-frame segmentation results
356
- # for out_frame_idx, out_obj_ids, out_mask_logits in predictor.propagate_in_video(inference_state):
357
- # video_segments[out_frame_idx] = {
358
- # out_obj_id: (out_mask_logits[i] > 0.0).cpu().numpy()
359
- # for i, out_obj_id in enumerate(out_obj_ids)
360
- # }
361
-
362
- out_obj_ids, out_mask_logits = predictor.propagate_in_video(inference_state, start_frame_idx=0, reverse=False,)
363
- print(out_obj_ids)
364
- for frame_idx in range(0, inference_state['num_frames']):
365
-
366
- video_segments[frame_idx] = {out_obj_ids[0]: (out_mask_logits[frame_idx]> 0.0).cpu().numpy()}
367
- # output_scores_per_object[object_id][frame_idx] = out_mask_logits[frame_idx].cpu().numpy()
368
-
369
- # render the segmentation results every few frames
370
- if vis_frame_type == "check":
371
- vis_frame_stride = 15
372
- elif vis_frame_type == "render":
373
- vis_frame_stride = 1
374
-
375
- plt.close("all")
376
- for out_frame_idx in range(0, len(frame_names), vis_frame_stride):
377
- plt.figure(figsize=(6, 4))
378
- plt.title(f"frame {out_frame_idx}")
379
- plt.imshow(Image.open(os.path.join(video_dir, frame_names[out_frame_idx])))
380
- for out_obj_id, out_mask in video_segments[out_frame_idx].items():
381
- show_mask(out_mask, plt.gca(), obj_id=out_obj_id)
382
-
383
- # Define the output filename and save the figure as a JPEG file
384
- output_filename = os.path.join(frames_output_dir, f"frame_{out_frame_idx}.jpg")
385
- plt.savefig(output_filename, format='jpg')
386
-
387
- # Close the plot
388
- plt.close()
389
-
390
- # Append the file path to the list
391
- jpeg_images.append(output_filename)
392
-
393
- if f"frame_{out_frame_idx}.jpg" not in available_frames_to_check:
394
- available_frames_to_check.append(f"frame_{out_frame_idx}.jpg")
395
-
396
- torch.cuda.empty_cache()
397
- print(f"JPEG_IMAGES: {jpeg_images}")
398
-
399
- if vis_frame_type == "check":
400
- return gr.update(value=jpeg_images), gr.update(value=None), gr.update(choices=available_frames_to_check, value=working_frame, visible=True), available_frames_to_check, gr.update(visible=True)
401
- elif vis_frame_type == "render":
402
- # Create a video clip from the image sequence
403
- original_fps = get_video_fps(video_in)
404
- fps = original_fps # Frames per second
405
- total_frames = len(jpeg_images)
406
- clip = ImageSequenceClip(jpeg_images, fps=fps)
407
- # Write the result to a file
408
- final_vid_output_path = "output_video.mp4"
409
-
410
- # Write the result to a file
411
- clip.write_videofile(
412
- final_vid_output_path,
413
- codec='libx264'
414
- )
415
-
416
- return gr.update(value=None), gr.update(value=final_vid_output_path), working_frame, available_frames_to_check, gr.update(visible=True)
417
-
418
- def update_ui(vis_frame_type):
419
- if vis_frame_type == "check":
420
- return gr.update(visible=True), gr.update(visible=False)
421
- elif vis_frame_type == "render":
422
- return gr.update(visible=False), gr.update(visible=True)
423
-
424
- def switch_working_frame(working_frame, scanned_frames, video_frames_dir):
425
- new_working_frame = None
426
- if working_frame == None:
427
- new_working_frame = os.path.join(video_frames_dir, scanned_frames[0])
428
-
429
- else:
430
- # Use a regular expression to find the integer
431
- match = re.search(r'frame_(\d+)', working_frame)
432
- if match:
433
- # Extract the integer from the match
434
- frame_number = int(match.group(1))
435
- ann_frame_idx = frame_number
436
- new_working_frame = os.path.join(video_frames_dir, scanned_frames[ann_frame_idx])
437
- return gr.State([]), gr.State([]), new_working_frame, new_working_frame
438
-
439
- def reset_propagation(first_frame_path, predictor, stored_inference_state):
440
-
441
- predictor.reset_state(stored_inference_state)
442
- # print(f"RESET State: {stored_inference_state} ")
443
- return first_frame_path, gr.State([]), gr.State([]), gr.update(value=None, visible=False), stored_inference_state, None, ["frame_0.jpg"], first_frame_path, "frame_0.jpg", gr.update(visible=False)
444
-
445
-
446
- with gr.Blocks(css=css) as demo:
447
- first_frame_path = gr.State()
448
- tracking_points = gr.State([])
449
- trackings_input_label = gr.State([])
450
- video_frames_dir = gr.State()
451
- scanned_frames = gr.State()
452
- loaded_predictor = gr.State()
453
- stored_inference_state = gr.State()
454
- stored_frame_names = gr.State()
455
- available_frames_to_check = gr.State([])
456
- with gr.Column():
457
- gr.Markdown(
458
- """
459
- <h1 style="text-align: center;">🔥 SAM2Long Demo 🔥</h1>
460
- """
461
- )
462
- gr.Markdown(
463
- """
464
- This is a simple demo for video segmentation with [SAM2Long](https://github.com/Mark12Ding/SAM2Long).
465
- """
466
- )
467
- gr.Markdown(
468
- """
469
- ### 📋 Instructions:
470
-
471
- It is largely built on the [SAM2-Video-Predictor](https://huggingface.co/spaces/fffiloni/SAM2-Video-Predictor).
472
-
473
- 1. **Upload your video** [MP4-24fps]
474
- 2. With **'include' point type** selected, click on the object to mask on the first frame
475
- 3. Switch to **'exclude' point type** if you want to specify an area to avoid
476
- 4. **Get Mask!**
477
- 5. **Check Propagation** every 15 frames
478
- 6. **Propagate with "render"** to render the final masked video
479
- 7. **Hit Reset** button if you want to refresh and start again
480
-
481
- *Note: Input video will be processed for up to 10 seconds only for demo purposes.*
482
- """
483
- )
484
- with gr.Row():
485
-
486
- with gr.Column():
487
- with gr.Group():
488
- with gr.Group():
489
- with gr.Row():
490
- point_type = gr.Radio(label="point type", choices=["include", "exclude"], value="include", scale=2)
491
- clear_points_btn = gr.Button("Clear Points", scale=1)
492
-
493
- input_first_frame_image = gr.Image(label="input image", interactive=False, type="filepath", visible=False)
494
-
495
- points_map = gr.Image(
496
- label="Point n Click map",
497
- type="filepath",
498
- interactive=False
499
- )
500
-
501
- with gr.Group():
502
- with gr.Row():
503
- checkpoint = gr.Dropdown(label="Checkpoint", choices=["tiny", "small", "base-plus"], value="tiny")
504
- submit_btn = gr.Button("Get Mask", size="lg")
505
-
506
- with gr.Accordion("Your video IN", open=True) as video_in_drawer:
507
- video_in = gr.Video(label="Video IN", format="mp4")
508
-
509
- gr.HTML("""
510
-
511
- <a href="https://huggingface.co/spaces/{os.environ['SPACE_ID']}?duplicate=true">
512
- <img src="https://huggingface.co/datasets/huggingface/badges/resolve/main/duplicate-this-space-lg-dark.svg" alt="Duplicate this Space" />
513
- </a> to skip queue and avoid OOM errors from heavy public load
514
- """)
515
-
516
- with gr.Column():
517
- with gr.Group():
518
- # with gr.Group():
519
- # with gr.Row():
520
- working_frame = gr.Dropdown(label="working frame ID", choices=[""], value="frame_0.jpg", visible=False, allow_custom_value=False, interactive=True)
521
- # change_current = gr.Button("change current", visible=False)
522
- # working_frame = []
523
- output_result = gr.Image(label="current working mask ref")
524
- with gr.Group():
525
- with gr.Row():
526
- vis_frame_type = gr.Radio(label="Propagation level", choices=["check", "render"], value="check", scale=2)
527
- propagate_btn = gr.Button("Propagate", scale=1)
528
- reset_prpgt_brn = gr.Button("Reset", visible=False)
529
- output_propagated = gr.Gallery(label="Propagated Mask samples gallery", columns=4, visible=False)
530
- output_video = gr.Video(visible=False)
531
- # output_result_mask = gr.Image()
532
-
533
-
534
-
535
- # When new video is uploaded
536
- video_in.upload(
537
- fn = preprocess_video_in,
538
- inputs = [video_in],
539
- outputs = [
540
- first_frame_path,
541
- tracking_points, # update Tracking Points in the gr.State([]) object
542
- trackings_input_label, # update Tracking Labels in the gr.State([]) object
543
- input_first_frame_image, # hidden component used as ref when clearing points
544
- points_map, # Image component where we add new tracking points
545
- video_frames_dir, # Array where frames from video_in are deep stored
546
- scanned_frames, # Scanned frames by SAM2
547
- stored_inference_state, # Sam2 inference state
548
- stored_frame_names, #
549
- video_in_drawer, # Accordion to hide uploaded video player
550
- ],
551
- queue = False
552
- )
553
-
554
-
555
- # triggered when we click on image to add new points
556
- points_map.select(
557
- fn = get_point,
558
- inputs = [
559
- point_type, # "include" or "exclude"
560
- tracking_points, # get tracking_points values
561
- trackings_input_label, # get tracking label values
562
- input_first_frame_image, # gr.State() first frame path
563
- ],
564
- outputs = [
565
- tracking_points, # updated with new points
566
- trackings_input_label, # updated with corresponding labels
567
- points_map, # updated image with points
568
- ],
569
- queue = False
570
- )
571
-
572
- # Clear every points clicked and added to the map
573
- clear_points_btn.click(
574
- fn = clear_points,
575
- inputs = input_first_frame_image, # we get the untouched hidden image
576
- outputs = [
577
- first_frame_path,
578
- tracking_points,
579
- trackings_input_label,
580
- points_map,
581
- #stored_inference_state,
582
- ],
583
- queue=False
584
- )
585
-
586
-
587
- # change_current.click(
588
- # fn = switch_working_frame,
589
- # inputs = [working_frame, scanned_frames, video_frames_dir],
590
- # outputs = [tracking_points, trackings_input_label, input_first_frame_image, points_map],
591
- # queue=False
592
- # )
593
-
594
-
595
- submit_btn.click(
596
- fn = get_mask_sam_process,
597
- inputs = [
598
- stored_inference_state,
599
- input_first_frame_image,
600
- checkpoint,
601
- tracking_points,
602
- trackings_input_label,
603
- video_frames_dir,
604
- scanned_frames,
605
- working_frame,
606
- available_frames_to_check,
607
  ],
608
- outputs = [
609
- output_result,
610
- stored_frame_names,
611
- loaded_predictor,
612
- stored_inference_state,
613
- working_frame,
614
- ],
615
- queue=False
616
- )
617
-
618
- reset_prpgt_brn.click(
619
- fn = reset_propagation,
620
- inputs = [first_frame_path, loaded_predictor, stored_inference_state],
621
- outputs = [points_map, tracking_points, trackings_input_label, output_propagated, stored_inference_state, output_result, available_frames_to_check, input_first_frame_image, working_frame, reset_prpgt_brn],
622
- queue=False
623
- )
624
-
625
- propagate_btn.click(
626
- fn = update_ui,
627
- inputs = [vis_frame_type],
628
- outputs = [output_propagated, output_video],
629
- queue=False
630
- ).then(
631
- fn = propagate_to_all,
632
- inputs = [video_in, checkpoint, stored_inference_state, stored_frame_names, video_frames_dir, vis_frame_type, available_frames_to_check, working_frame],
633
- outputs = [output_propagated, output_video, working_frame, available_frames_to_check, reset_prpgt_brn]
634
- )
635
-
636
- demo.queue().launch(show_api=False, show_error=True, share=True, server_name="0.0.0.0", server_port=11111)
 
1
+ # Copyright (c) Meta Platforms, Inc. and affiliates.
2
+ # All rights reserved.
3
+
4
+ # This source code is licensed under the license found in the
5
+ # LICENSE file in the root directory of this source tree.
6
+
7
+ from setuptools import find_packages, setup
8
+ from torch.utils.cpp_extension import BuildExtension, CUDAExtension
9
+
10
+ # Package metadata
11
+ NAME = "SAM 2"
12
+ VERSION = "1.0"
13
+ DESCRIPTION = "SAM 2: Segment Anything in Images and Videos"
14
+ URL = "https://github.com/facebookresearch/segment-anything-2"
15
+ AUTHOR = "Meta AI"
16
+ AUTHOR_EMAIL = "segment-anything@meta.com"
17
+ LICENSE = "Apache 2.0"
18
+
19
+ # Read the contents of README file
20
+ with open("README.md", "r") as f:
21
+ LONG_DESCRIPTION = f.read()
22
+
23
+ # Required dependencies
24
+ REQUIRED_PACKAGES = [
25
+ "torch>=2.3.1",
26
+ "torchvision>=0.18.1",
27
+ "numpy>=1.24.4",
28
+ "tqdm>=4.66.1",
29
+ "hydra-core>=1.3.2",
30
+ "iopath>=0.1.10",
31
+ "pillow>=9.4.0",
32
+ ]
33
+
34
+ EXTRA_PACKAGES = {
35
+ "demo": ["matplotlib>=3.9.1", "jupyter>=1.0.0", "opencv-python>=4.7.0"],
36
+ "dev": ["black==24.2.0", "usort==1.0.2", "ufmt==2.0.0b2"],
 
37
  }
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
38
 
 
 
 
 
 
 
39
 
40
+ def get_extensions():
41
+ srcs = ["sam2/csrc/connected_components.cu"]
42
+ compile_args = {
43
+ "cxx": [],
44
+ "nvcc": [
45
+ "-DCUDA_HAS_FP16=1",
46
+ "-D__CUDA_NO_HALF_OPERATORS__",
47
+ "-D__CUDA_NO_HALF_CONVERSIONS__",
48
+ "-D__CUDA_NO_HALF2_OPERATORS__",
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
49
  ],
50
+ }
51
+ ext_modules = [CUDAExtension("sam2._C", srcs, extra_compile_args=compile_args)]
52
+ return ext_modules
53
+
54
+
55
+ # Setup configuration
56
+ setup(
57
+ name=NAME,
58
+ version=VERSION,
59
+ description=DESCRIPTION,
60
+ long_description=LONG_DESCRIPTION,
61
+ long_description_content_type="text/markdown",
62
+ url=URL,
63
+ author=AUTHOR,
64
+ author_email=AUTHOR_EMAIL,
65
+ license=LICENSE,
66
+ packages=find_packages(exclude="notebooks"),
67
+ install_requires=REQUIRED_PACKAGES,
68
+ extras_require=EXTRA_PACKAGES,
69
+ python_requires=">=3.10.0",
70
+ ext_modules=get_extensions(),
71
+ cmdclass={"build_ext": BuildExtension.with_options(no_python_abi_suffix=True)},
72
+ )