update for line plot frontend
Browse files- api.py +703 -529
- requirements.txt +22 -24
api.py
CHANGED
@@ -1,530 +1,704 @@
|
|
1 |
-
from fastapi import FastAPI, HTTPException, UploadFile, File, Form
|
2 |
-
from fastapi.middleware.cors import CORSMiddleware
|
3 |
-
from pydantic import BaseModel
|
4 |
-
from typing import Dict, List, Any, Optional
|
5 |
-
import json
|
6 |
-
import tempfile
|
7 |
-
import os
|
8 |
-
from PIL import Image
|
9 |
-
import numpy as np
|
10 |
-
import cv2
|
11 |
-
import torch
|
12 |
-
import torchvision.transforms as T
|
13 |
-
import torchvision.transforms.functional as f
|
14 |
-
import yaml
|
15 |
-
from tqdm import tqdm
|
16 |
-
from huggingface_hub import hf_hub_download
|
17 |
-
|
18 |
-
from get_camera_params import get_camera_parameters
|
19 |
-
|
20 |
-
# Imports pour l'inférence automatique
|
21 |
-
from model.cls_hrnet import get_cls_net
|
22 |
-
from model.cls_hrnet_l import get_cls_net as get_cls_net_l
|
23 |
-
from utils.utils_calib import FramebyFrameCalib
|
24 |
-
from utils.utils_heatmap import get_keypoints_from_heatmap_batch_maxpool, get_keypoints_from_heatmap_batch_maxpool_l, complete_keypoints, coords_to_dict
|
25 |
-
|
26 |
-
app = FastAPI(
|
27 |
-
title="Football Vision Calibration API",
|
28 |
-
description="API pour la calibration de caméras à partir de lignes de terrain de football",
|
29 |
-
version="1.0.0"
|
30 |
-
)
|
31 |
-
|
32 |
-
# Configuration CORS pour autoriser les requêtes depuis le frontend
|
33 |
-
app.add_middleware(
|
34 |
-
CORSMiddleware,
|
35 |
-
allow_origins=["*"], # En production, spécifiez les domaines autorisés
|
36 |
-
allow_credentials=True,
|
37 |
-
allow_methods=["*"],
|
38 |
-
allow_headers=["*"],
|
39 |
-
)
|
40 |
-
|
41 |
-
# Paramètres par défaut pour l'inférence
|
42 |
-
WEIGHTS_KP = "models/SV_FT_TSWC_kp"
|
43 |
-
WEIGHTS_LINE = "models/SV_FT_TSWC_lines"
|
44 |
-
# DEVICE = "cuda:0"
|
45 |
-
DEVICE = "cpu"
|
46 |
-
KP_THRESHOLD = 0.15
|
47 |
-
LINE_THRESHOLD = 0.15
|
48 |
-
PNL_REFINE = True
|
49 |
-
FRAME_STEP = 5
|
50 |
-
|
51 |
-
# Cache pour les modèles (éviter de les recharger à chaque requête)
|
52 |
-
_models_cache = None
|
53 |
-
|
54 |
-
# Paramètres pour HF Hub
|
55 |
-
HF_MODEL_REPO = "2nzi/SV_FT_TSWC_kp" # Remplacez par votre repo
|
56 |
-
WEIGHTS_KP_FILE = "SV_FT_TSWC_kp" # Nom du fichier dans le repo
|
57 |
-
WEIGHTS_LINE_FILE = "SV_FT_TSWC_lines" # Nom du fichier dans le repo
|
58 |
-
|
59 |
-
def load_inference_models():
|
60 |
-
"""Charge les modèles d'inférence depuis Hugging Face Hub"""
|
61 |
-
global _models_cache
|
62 |
-
|
63 |
-
if _models_cache is not None:
|
64 |
-
return _models_cache
|
65 |
-
|
66 |
-
try:
|
67 |
-
# Device detection
|
68 |
-
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
|
69 |
-
print(f"Using device: {device}")
|
70 |
-
|
71 |
-
# Télécharger les modèles depuis HF Hub
|
72 |
-
print("Téléchargement des modèles depuis Hugging Face Hub...")
|
73 |
-
|
74 |
-
weights_kp_path = hf_hub_download(
|
75 |
-
repo_id=HF_MODEL_REPO,
|
76 |
-
filename=WEIGHTS_KP_FILE,
|
77 |
-
cache_dir="./hf_cache"
|
78 |
-
)
|
79 |
-
|
80 |
-
weights_line_path = hf_hub_download(
|
81 |
-
repo_id=HF_MODEL_REPO,
|
82 |
-
filename=WEIGHTS_LINE_FILE,
|
83 |
-
cache_dir="./hf_cache"
|
84 |
-
)
|
85 |
-
|
86 |
-
print(f"Modèles téléchargés:")
|
87 |
-
print(f" - Keypoints: {weights_kp_path}")
|
88 |
-
print(f" - Lines: {weights_line_path}")
|
89 |
-
|
90 |
-
# Vérifier l'existence des fichiers de configuration
|
91 |
-
config_files = ["config/hrnetv2_w48.yaml", "config/hrnetv2_w48_l.yaml"]
|
92 |
-
for config_file in config_files:
|
93 |
-
if not os.path.exists(config_file):
|
94 |
-
raise FileNotFoundError(f"Fichier de configuration manquant: {config_file}")
|
95 |
-
|
96 |
-
# Charger les configurations
|
97 |
-
with open("config/hrnetv2_w48.yaml", 'r') as f:
|
98 |
-
cfg = yaml.safe_load(f)
|
99 |
-
with open("config/hrnetv2_w48_l.yaml", 'r') as f:
|
100 |
-
cfg_l = yaml.safe_load(f)
|
101 |
-
|
102 |
-
# Modèle keypoints
|
103 |
-
model = get_cls_net(cfg)
|
104 |
-
model.load_state_dict(torch.load(weights_kp_path, map_location=device))
|
105 |
-
model.to(device)
|
106 |
-
model.eval()
|
107 |
-
|
108 |
-
# Modèle lignes
|
109 |
-
model_l = get_cls_net_l(cfg_l)
|
110 |
-
model_l.load_state_dict(torch.load(weights_line_path, map_location=device))
|
111 |
-
model_l.to(device)
|
112 |
-
model_l.eval()
|
113 |
-
|
114 |
-
_models_cache = (model, model_l, device)
|
115 |
-
print("✅ Modèles chargés avec succès depuis HF Hub!")
|
116 |
-
return _models_cache
|
117 |
-
|
118 |
-
except Exception as e:
|
119 |
-
print(f"❌ Erreur lors du chargement des modèles: {e}")
|
120 |
-
raise HTTPException(
|
121 |
-
status_code=503,
|
122 |
-
detail=f"Modèles non disponibles: {str(e)}. Veuillez réessayer plus tard."
|
123 |
-
)
|
124 |
-
|
125 |
-
def process_frame_inference(frame, model, model_l, device, frame_width, frame_height):
|
126 |
-
"""Traite une frame et retourne les paramètres de caméra"""
|
127 |
-
transform = T.Resize((540, 960))
|
128 |
-
|
129 |
-
# Préparer la frame pour l'inférence
|
130 |
-
frame_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
|
131 |
-
frame_pil = Image.fromarray(frame_rgb)
|
132 |
-
frame_tensor = f.to_tensor(frame_pil).float().unsqueeze(0)
|
133 |
-
|
134 |
-
if frame_tensor.size()[-1] != 960:
|
135 |
-
frame_tensor = transform(frame_tensor)
|
136 |
-
|
137 |
-
frame_tensor = frame_tensor.to(device)
|
138 |
-
b, c, h, w = frame_tensor.size()
|
139 |
-
|
140 |
-
# Inférence
|
141 |
-
with torch.no_grad():
|
142 |
-
heatmaps = model(frame_tensor)
|
143 |
-
heatmaps_l = model_l(frame_tensor)
|
144 |
-
|
145 |
-
# Extraire les keypoints et lignes
|
146 |
-
kp_coords = get_keypoints_from_heatmap_batch_maxpool(heatmaps[:,:-1,:,:])
|
147 |
-
line_coords = get_keypoints_from_heatmap_batch_maxpool_l(heatmaps_l[:,:-1,:,:])
|
148 |
-
kp_dict = coords_to_dict(kp_coords, threshold=KP_THRESHOLD)
|
149 |
-
lines_dict = coords_to_dict(line_coords, threshold=LINE_THRESHOLD)
|
150 |
-
kp_dict, lines_dict = complete_keypoints(kp_dict[0], lines_dict[0], w=w, h=h, normalize=True)
|
151 |
-
|
152 |
-
# Calibration
|
153 |
-
cam = FramebyFrameCalib(iwidth=frame_width, iheight=frame_height, denormalize=True)
|
154 |
-
cam.update(kp_dict, lines_dict)
|
155 |
-
final_params_dict = cam.heuristic_voting(refine_lines=PNL_REFINE)
|
156 |
-
|
157 |
-
return final_params_dict
|
158 |
-
|
159 |
-
# Modèles Pydantic pour la validation des données
|
160 |
-
class Point(BaseModel):
|
161 |
-
x: float
|
162 |
-
y: float
|
163 |
-
|
164 |
-
class LinePolygon(BaseModel):
|
165 |
-
points: List[Point]
|
166 |
-
|
167 |
-
class CalibrationRequest(BaseModel):
|
168 |
-
lines: Dict[str, List[Point]]
|
169 |
-
|
170 |
-
|
171 |
-
|
172 |
-
|
173 |
-
|
174 |
-
|
175 |
-
|
176 |
-
|
177 |
-
|
178 |
-
|
179 |
-
|
180 |
-
|
181 |
-
|
182 |
-
|
183 |
-
|
184 |
-
|
185 |
-
|
186 |
-
|
187 |
-
|
188 |
-
|
189 |
-
|
190 |
-
|
191 |
-
|
192 |
-
|
193 |
-
|
194 |
-
|
195 |
-
|
196 |
-
|
197 |
-
|
198 |
-
|
199 |
-
|
200 |
-
|
201 |
-
|
202 |
-
|
203 |
-
|
204 |
-
|
205 |
-
|
206 |
-
|
207 |
-
|
208 |
-
|
209 |
-
|
210 |
-
|
211 |
-
|
212 |
-
|
213 |
-
|
214 |
-
|
215 |
-
|
216 |
-
|
217 |
-
|
218 |
-
|
219 |
-
|
220 |
-
|
221 |
-
|
222 |
-
|
223 |
-
|
224 |
-
|
225 |
-
|
226 |
-
|
227 |
-
|
228 |
-
|
229 |
-
|
230 |
-
|
231 |
-
|
232 |
-
|
233 |
-
|
234 |
-
|
235 |
-
|
236 |
-
|
237 |
-
|
238 |
-
|
239 |
-
|
240 |
-
|
241 |
-
|
242 |
-
|
243 |
-
|
244 |
-
#
|
245 |
-
|
246 |
-
|
247 |
-
|
248 |
-
|
249 |
-
|
250 |
-
|
251 |
-
|
252 |
-
|
253 |
-
|
254 |
-
|
255 |
-
|
256 |
-
|
257 |
-
|
258 |
-
|
259 |
-
|
260 |
-
|
261 |
-
|
262 |
-
|
263 |
-
|
264 |
-
|
265 |
-
|
266 |
-
|
267 |
-
|
268 |
-
|
269 |
-
|
270 |
-
|
271 |
-
|
272 |
-
|
273 |
-
|
274 |
-
|
275 |
-
|
276 |
-
|
277 |
-
|
278 |
-
|
279 |
-
|
280 |
-
|
281 |
-
|
282 |
-
|
283 |
-
|
284 |
-
|
285 |
-
|
286 |
-
|
287 |
-
|
288 |
-
#
|
289 |
-
|
290 |
-
|
291 |
-
|
292 |
-
|
293 |
-
|
294 |
-
|
295 |
-
|
296 |
-
|
297 |
-
|
298 |
-
|
299 |
-
|
300 |
-
|
301 |
-
|
302 |
-
|
303 |
-
|
304 |
-
|
305 |
-
|
306 |
-
|
307 |
-
|
308 |
-
|
309 |
-
|
310 |
-
|
311 |
-
|
312 |
-
|
313 |
-
|
314 |
-
|
315 |
-
|
316 |
-
|
317 |
-
|
318 |
-
|
319 |
-
|
320 |
-
|
321 |
-
|
322 |
-
|
323 |
-
|
324 |
-
|
325 |
-
|
326 |
-
|
327 |
-
|
328 |
-
|
329 |
-
|
330 |
-
|
331 |
-
|
332 |
-
|
333 |
-
|
334 |
-
|
335 |
-
|
336 |
-
|
337 |
-
|
338 |
-
|
339 |
-
|
340 |
-
|
341 |
-
|
342 |
-
|
343 |
-
|
344 |
-
|
345 |
-
|
346 |
-
|
347 |
-
|
348 |
-
|
349 |
-
|
350 |
-
|
351 |
-
|
352 |
-
|
353 |
-
|
354 |
-
|
355 |
-
|
356 |
-
|
357 |
-
|
358 |
-
|
359 |
-
|
360 |
-
|
361 |
-
|
362 |
-
|
363 |
-
|
364 |
-
|
365 |
-
|
366 |
-
|
367 |
-
|
368 |
-
|
369 |
-
|
370 |
-
|
371 |
-
|
372 |
-
|
373 |
-
|
374 |
-
|
375 |
-
|
376 |
-
|
377 |
-
|
378 |
-
|
379 |
-
|
380 |
-
|
381 |
-
|
382 |
-
|
383 |
-
|
384 |
-
|
385 |
-
|
386 |
-
|
387 |
-
|
388 |
-
|
389 |
-
|
390 |
-
|
391 |
-
|
392 |
-
|
393 |
-
|
394 |
-
|
395 |
-
|
396 |
-
|
397 |
-
|
398 |
-
|
399 |
-
|
400 |
-
|
401 |
-
|
402 |
-
|
403 |
-
|
404 |
-
|
405 |
-
|
406 |
-
|
407 |
-
|
408 |
-
|
409 |
-
|
410 |
-
|
411 |
-
|
412 |
-
|
413 |
-
|
414 |
-
|
415 |
-
|
416 |
-
|
417 |
-
|
418 |
-
|
419 |
-
|
420 |
-
|
421 |
-
|
422 |
-
|
423 |
-
|
424 |
-
|
425 |
-
|
426 |
-
|
427 |
-
|
428 |
-
|
429 |
-
|
430 |
-
|
431 |
-
|
432 |
-
|
433 |
-
|
434 |
-
|
435 |
-
|
436 |
-
|
437 |
-
|
438 |
-
|
439 |
-
|
440 |
-
|
441 |
-
|
442 |
-
|
443 |
-
|
444 |
-
|
445 |
-
|
446 |
-
|
447 |
-
|
448 |
-
|
449 |
-
|
450 |
-
|
451 |
-
|
452 |
-
|
453 |
-
|
454 |
-
|
455 |
-
|
456 |
-
|
457 |
-
|
458 |
-
|
459 |
-
|
460 |
-
|
461 |
-
|
462 |
-
|
463 |
-
|
464 |
-
|
465 |
-
|
466 |
-
|
467 |
-
|
468 |
-
|
469 |
-
|
470 |
-
|
471 |
-
|
472 |
-
|
473 |
-
|
474 |
-
|
475 |
-
|
476 |
-
|
477 |
-
|
478 |
-
|
479 |
-
|
480 |
-
|
481 |
-
|
482 |
-
|
483 |
-
|
484 |
-
|
485 |
-
|
486 |
-
|
487 |
-
|
488 |
-
|
489 |
-
|
490 |
-
|
491 |
-
|
492 |
-
|
493 |
-
|
494 |
-
|
495 |
-
|
496 |
-
|
497 |
-
|
498 |
-
|
499 |
-
|
500 |
-
|
501 |
-
|
502 |
-
|
503 |
-
|
504 |
-
|
505 |
-
|
506 |
-
|
507 |
-
|
508 |
-
|
509 |
-
|
510 |
-
|
511 |
-
|
512 |
-
|
513 |
-
|
514 |
-
|
515 |
-
|
516 |
-
|
517 |
-
|
518 |
-
|
519 |
-
|
520 |
-
|
521 |
-
|
522 |
-
|
523 |
-
|
524 |
-
|
525 |
-
|
526 |
-
|
527 |
-
|
528 |
-
|
529 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
530 |
app_instance = app
|
|
|
1 |
+
from fastapi import FastAPI, HTTPException, UploadFile, File, Form
|
2 |
+
from fastapi.middleware.cors import CORSMiddleware
|
3 |
+
from pydantic import BaseModel
|
4 |
+
from typing import Dict, List, Any, Optional
|
5 |
+
import json
|
6 |
+
import tempfile
|
7 |
+
import os
|
8 |
+
from PIL import Image
|
9 |
+
import numpy as np
|
10 |
+
import cv2
|
11 |
+
import torch
|
12 |
+
import torchvision.transforms as T
|
13 |
+
import torchvision.transforms.functional as f
|
14 |
+
import yaml
|
15 |
+
from tqdm import tqdm
|
16 |
+
from huggingface_hub import hf_hub_download
|
17 |
+
|
18 |
+
from get_camera_params import get_camera_parameters
|
19 |
+
|
20 |
+
# Imports pour l'inférence automatique
|
21 |
+
from model.cls_hrnet import get_cls_net
|
22 |
+
from model.cls_hrnet_l import get_cls_net as get_cls_net_l
|
23 |
+
from utils.utils_calib import FramebyFrameCalib
|
24 |
+
from utils.utils_heatmap import get_keypoints_from_heatmap_batch_maxpool, get_keypoints_from_heatmap_batch_maxpool_l, complete_keypoints, coords_to_dict
|
25 |
+
|
26 |
+
app = FastAPI(
|
27 |
+
title="Football Vision Calibration API",
|
28 |
+
description="API pour la calibration de caméras à partir de lignes de terrain de football",
|
29 |
+
version="1.0.0"
|
30 |
+
)
|
31 |
+
|
32 |
+
# Configuration CORS pour autoriser les requêtes depuis le frontend
|
33 |
+
app.add_middleware(
|
34 |
+
CORSMiddleware,
|
35 |
+
allow_origins=["*"], # En production, spécifiez les domaines autorisés
|
36 |
+
allow_credentials=True,
|
37 |
+
allow_methods=["*"],
|
38 |
+
allow_headers=["*"],
|
39 |
+
)
|
40 |
+
|
41 |
+
# Paramètres par défaut pour l'inférence
|
42 |
+
WEIGHTS_KP = "models/SV_FT_TSWC_kp"
|
43 |
+
WEIGHTS_LINE = "models/SV_FT_TSWC_lines"
|
44 |
+
# DEVICE = "cuda:0"
|
45 |
+
DEVICE = "cpu"
|
46 |
+
KP_THRESHOLD = 0.15
|
47 |
+
LINE_THRESHOLD = 0.15
|
48 |
+
PNL_REFINE = True
|
49 |
+
FRAME_STEP = 5
|
50 |
+
|
51 |
+
# Cache pour les modèles (éviter de les recharger à chaque requête)
|
52 |
+
_models_cache = None
|
53 |
+
|
54 |
+
# Paramètres pour HF Hub
|
55 |
+
HF_MODEL_REPO = "2nzi/SV_FT_TSWC_kp" # Remplacez par votre repo
|
56 |
+
WEIGHTS_KP_FILE = "SV_FT_TSWC_kp" # Nom du fichier dans le repo
|
57 |
+
WEIGHTS_LINE_FILE = "SV_FT_TSWC_lines" # Nom du fichier dans le repo
|
58 |
+
|
59 |
+
def load_inference_models():
|
60 |
+
"""Charge les modèles d'inférence depuis Hugging Face Hub"""
|
61 |
+
global _models_cache
|
62 |
+
|
63 |
+
if _models_cache is not None:
|
64 |
+
return _models_cache
|
65 |
+
|
66 |
+
try:
|
67 |
+
# Device detection
|
68 |
+
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
|
69 |
+
print(f"Using device: {device}")
|
70 |
+
|
71 |
+
# Télécharger les modèles depuis HF Hub
|
72 |
+
print("Téléchargement des modèles depuis Hugging Face Hub...")
|
73 |
+
|
74 |
+
weights_kp_path = hf_hub_download(
|
75 |
+
repo_id=HF_MODEL_REPO,
|
76 |
+
filename=WEIGHTS_KP_FILE,
|
77 |
+
cache_dir="./hf_cache"
|
78 |
+
)
|
79 |
+
|
80 |
+
weights_line_path = hf_hub_download(
|
81 |
+
repo_id=HF_MODEL_REPO,
|
82 |
+
filename=WEIGHTS_LINE_FILE,
|
83 |
+
cache_dir="./hf_cache"
|
84 |
+
)
|
85 |
+
|
86 |
+
print(f"Modèles téléchargés:")
|
87 |
+
print(f" - Keypoints: {weights_kp_path}")
|
88 |
+
print(f" - Lines: {weights_line_path}")
|
89 |
+
|
90 |
+
# Vérifier l'existence des fichiers de configuration
|
91 |
+
config_files = ["config/hrnetv2_w48.yaml", "config/hrnetv2_w48_l.yaml"]
|
92 |
+
for config_file in config_files:
|
93 |
+
if not os.path.exists(config_file):
|
94 |
+
raise FileNotFoundError(f"Fichier de configuration manquant: {config_file}")
|
95 |
+
|
96 |
+
# Charger les configurations
|
97 |
+
with open("config/hrnetv2_w48.yaml", 'r') as f:
|
98 |
+
cfg = yaml.safe_load(f)
|
99 |
+
with open("config/hrnetv2_w48_l.yaml", 'r') as f:
|
100 |
+
cfg_l = yaml.safe_load(f)
|
101 |
+
|
102 |
+
# Modèle keypoints
|
103 |
+
model = get_cls_net(cfg)
|
104 |
+
model.load_state_dict(torch.load(weights_kp_path, map_location=device))
|
105 |
+
model.to(device)
|
106 |
+
model.eval()
|
107 |
+
|
108 |
+
# Modèle lignes
|
109 |
+
model_l = get_cls_net_l(cfg_l)
|
110 |
+
model_l.load_state_dict(torch.load(weights_line_path, map_location=device))
|
111 |
+
model_l.to(device)
|
112 |
+
model_l.eval()
|
113 |
+
|
114 |
+
_models_cache = (model, model_l, device)
|
115 |
+
print("✅ Modèles chargés avec succès depuis HF Hub!")
|
116 |
+
return _models_cache
|
117 |
+
|
118 |
+
except Exception as e:
|
119 |
+
print(f"❌ Erreur lors du chargement des modèles: {e}")
|
120 |
+
raise HTTPException(
|
121 |
+
status_code=503,
|
122 |
+
detail=f"Modèles non disponibles: {str(e)}. Veuillez réessayer plus tard."
|
123 |
+
)
|
124 |
+
|
125 |
+
def process_frame_inference(frame, model, model_l, device, frame_width, frame_height):
|
126 |
+
"""Traite une frame et retourne les paramètres de caméra"""
|
127 |
+
transform = T.Resize((540, 960))
|
128 |
+
|
129 |
+
# Préparer la frame pour l'inférence
|
130 |
+
frame_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
|
131 |
+
frame_pil = Image.fromarray(frame_rgb)
|
132 |
+
frame_tensor = f.to_tensor(frame_pil).float().unsqueeze(0)
|
133 |
+
|
134 |
+
if frame_tensor.size()[-1] != 960:
|
135 |
+
frame_tensor = transform(frame_tensor)
|
136 |
+
|
137 |
+
frame_tensor = frame_tensor.to(device)
|
138 |
+
b, c, h, w = frame_tensor.size()
|
139 |
+
|
140 |
+
# Inférence
|
141 |
+
with torch.no_grad():
|
142 |
+
heatmaps = model(frame_tensor)
|
143 |
+
heatmaps_l = model_l(frame_tensor)
|
144 |
+
|
145 |
+
# Extraire les keypoints et lignes
|
146 |
+
kp_coords = get_keypoints_from_heatmap_batch_maxpool(heatmaps[:,:-1,:,:])
|
147 |
+
line_coords = get_keypoints_from_heatmap_batch_maxpool_l(heatmaps_l[:,:-1,:,:])
|
148 |
+
kp_dict = coords_to_dict(kp_coords, threshold=KP_THRESHOLD)
|
149 |
+
lines_dict = coords_to_dict(line_coords, threshold=LINE_THRESHOLD)
|
150 |
+
kp_dict, lines_dict = complete_keypoints(kp_dict[0], lines_dict[0], w=w, h=h, normalize=True)
|
151 |
+
|
152 |
+
# Calibration
|
153 |
+
cam = FramebyFrameCalib(iwidth=frame_width, iheight=frame_height, denormalize=True)
|
154 |
+
cam.update(kp_dict, lines_dict)
|
155 |
+
final_params_dict = cam.heuristic_voting(refine_lines=PNL_REFINE)
|
156 |
+
|
157 |
+
return final_params_dict
|
158 |
+
|
159 |
+
# Modèles Pydantic pour la validation des données
|
160 |
+
class Point(BaseModel):
|
161 |
+
x: float
|
162 |
+
y: float
|
163 |
+
|
164 |
+
class LinePolygon(BaseModel):
|
165 |
+
points: List[Point]
|
166 |
+
|
167 |
+
class CalibrationRequest(BaseModel):
|
168 |
+
lines: Dict[str, List[Point]]
|
169 |
+
|
170 |
+
# Nouvelles classes pour les coordonnées
|
171 |
+
class ImageCoordinate(BaseModel):
|
172 |
+
x: float
|
173 |
+
y: float
|
174 |
+
|
175 |
+
class WorldCoordinate(BaseModel):
|
176 |
+
x: float # Coordonnée X sur le terrain (mètres)
|
177 |
+
y: float # Coordonnée Y sur le terrain (mètres)
|
178 |
+
|
179 |
+
class KeypointData(BaseModel):
|
180 |
+
id: int
|
181 |
+
image_coords: ImageCoordinate
|
182 |
+
world_coords: Optional[WorldCoordinate] = None
|
183 |
+
confidence: Optional[float] = None
|
184 |
+
|
185 |
+
class LineData(BaseModel):
|
186 |
+
id: int
|
187 |
+
start_image: ImageCoordinate
|
188 |
+
end_image: ImageCoordinate
|
189 |
+
start_world: Optional[WorldCoordinate] = None
|
190 |
+
end_world: Optional[WorldCoordinate] = None
|
191 |
+
confidence: Optional[float] = None
|
192 |
+
|
193 |
+
class DetectionData(BaseModel):
|
194 |
+
keypoints: List[KeypointData]
|
195 |
+
lines: List[LineData]
|
196 |
+
|
197 |
+
class CalibrationResponse(BaseModel):
|
198 |
+
status: str
|
199 |
+
camera_parameters: Dict[str, Any]
|
200 |
+
input_lines: Dict[str, List[Point]]
|
201 |
+
detections: Optional[DetectionData] = None
|
202 |
+
message: str
|
203 |
+
|
204 |
+
class InferenceImageResponse(BaseModel):
|
205 |
+
status: str
|
206 |
+
camera_parameters: Optional[Dict[str, Any]]
|
207 |
+
image_info: Dict[str, Any]
|
208 |
+
detections: Optional[DetectionData] = None
|
209 |
+
message: str
|
210 |
+
|
211 |
+
class FrameResult(BaseModel):
|
212 |
+
frame_number: int
|
213 |
+
timestamp_seconds: float
|
214 |
+
camera_parameters: Optional[Dict[str, Any]]
|
215 |
+
detections: Optional[DetectionData] = None
|
216 |
+
|
217 |
+
class InferenceVideoResponse(BaseModel):
|
218 |
+
status: str
|
219 |
+
video_info: Dict[str, Any]
|
220 |
+
frames_processed: int
|
221 |
+
frames_results: List[FrameResult]
|
222 |
+
message: str
|
223 |
+
|
224 |
+
# Fonction de conversion coordonnées image -> terrain
|
225 |
+
def image_to_world(point_2d, cam_params):
|
226 |
+
"""
|
227 |
+
Convertit un point 2D de l'image vers des coordonnées 3D du terrain.
|
228 |
+
|
229 |
+
Args:
|
230 |
+
point_2d: [x, y] coordonnées dans l'image
|
231 |
+
cam_params: Paramètres de la caméra
|
232 |
+
|
233 |
+
Returns:
|
234 |
+
[x, y] coordonnées sur le terrain (Z=0)
|
235 |
+
"""
|
236 |
+
try:
|
237 |
+
# Matrice de calibration
|
238 |
+
K = np.array([
|
239 |
+
[cam_params["cam_params"]["x_focal_length"], 0, cam_params["cam_params"]["principal_point"][0]],
|
240 |
+
[0, cam_params["cam_params"]["y_focal_length"], cam_params["cam_params"]["principal_point"][1]],
|
241 |
+
[0, 0, 1]
|
242 |
+
])
|
243 |
+
|
244 |
+
# Matrice de rotation et position
|
245 |
+
R = np.array(cam_params["cam_params"]["rotation_matrix"])
|
246 |
+
camera_pos = np.array(cam_params["cam_params"]["position_meters"])
|
247 |
+
|
248 |
+
# Point 2D en coordonnées homogènes
|
249 |
+
point_2d_h = np.array([point_2d[0], point_2d[1], 1])
|
250 |
+
|
251 |
+
# Back-projection du rayon depuis la caméra
|
252 |
+
ray = np.linalg.inv(K) @ point_2d_h
|
253 |
+
ray = R.T @ ray
|
254 |
+
|
255 |
+
# Intersection avec le plan Z=0 (terrain)
|
256 |
+
if abs(ray[2]) < 1e-6: # Éviter division par zéro
|
257 |
+
return None
|
258 |
+
|
259 |
+
t = -camera_pos[2] / ray[2]
|
260 |
+
world_point = camera_pos + t * ray
|
261 |
+
|
262 |
+
return world_point[:2] # Retourner seulement X,Y (Z=0)
|
263 |
+
|
264 |
+
except Exception as e:
|
265 |
+
print(f"Erreur conversion image->monde: {e}")
|
266 |
+
return None
|
267 |
+
|
268 |
+
def process_detections(kp_dict, lines_dict, cam_params):
|
269 |
+
"""
|
270 |
+
Traite les détections et convertit les coordonnées image vers terrain.
|
271 |
+
|
272 |
+
Args:
|
273 |
+
kp_dict: Dictionnaire des keypoints détectés
|
274 |
+
lines_dict: Dictionnaire des lignes détectées
|
275 |
+
cam_params: Paramètres de la caméra
|
276 |
+
|
277 |
+
Returns:
|
278 |
+
DetectionData: Données structurées avec coordonnées image et terrain
|
279 |
+
"""
|
280 |
+
keypoints_data = []
|
281 |
+
lines_data = []
|
282 |
+
|
283 |
+
# Traiter les keypoints
|
284 |
+
if kp_dict:
|
285 |
+
for kp_id, kp_value in kp_dict.items():
|
286 |
+
image_coords = ImageCoordinate(x=kp_value['x'], y=kp_value['y'])
|
287 |
+
|
288 |
+
# Convertir vers coordonnées terrain
|
289 |
+
world_coords = None
|
290 |
+
if cam_params and 'cam_params' in cam_params:
|
291 |
+
world_point = image_to_world([kp_value['x'], kp_value['y']], cam_params)
|
292 |
+
if world_point is not None:
|
293 |
+
world_coords = WorldCoordinate(x=float(world_point[0]), y=float(world_point[1]))
|
294 |
+
|
295 |
+
keypoints_data.append(KeypointData(
|
296 |
+
id=kp_id,
|
297 |
+
image_coords=image_coords,
|
298 |
+
world_coords=world_coords,
|
299 |
+
confidence=kp_value.get('confidence', None)
|
300 |
+
))
|
301 |
+
|
302 |
+
# Traiter les lignes
|
303 |
+
if lines_dict:
|
304 |
+
for line_id, line_value in lines_dict.items():
|
305 |
+
start_image = ImageCoordinate(x=line_value['x_1'], y=line_value['y_1'])
|
306 |
+
end_image = ImageCoordinate(x=line_value['x_2'], y=line_value['y_2'])
|
307 |
+
|
308 |
+
# Convertir vers coordonnées terrain
|
309 |
+
start_world = None
|
310 |
+
end_world = None
|
311 |
+
if cam_params and 'cam_params' in cam_params:
|
312 |
+
start_point = image_to_world([line_value['x_1'], line_value['y_1']], cam_params)
|
313 |
+
end_point = image_to_world([line_value['x_2'], line_value['y_2']], cam_params)
|
314 |
+
|
315 |
+
if start_point is not None:
|
316 |
+
start_world = WorldCoordinate(x=float(start_point[0]), y=float(start_point[1]))
|
317 |
+
if end_point is not None:
|
318 |
+
end_world = WorldCoordinate(x=float(end_point[0]), y=float(end_point[1]))
|
319 |
+
|
320 |
+
lines_data.append(LineData(
|
321 |
+
id=line_id,
|
322 |
+
start_image=start_image,
|
323 |
+
end_image=end_image,
|
324 |
+
start_world=start_world,
|
325 |
+
end_world=end_world,
|
326 |
+
confidence=line_value.get('confidence', None)
|
327 |
+
))
|
328 |
+
|
329 |
+
return DetectionData(keypoints=keypoints_data, lines=lines_data)
|
330 |
+
|
331 |
+
def process_frame_inference_with_coords(frame, model, model_l, device, frame_width, frame_height):
|
332 |
+
"""
|
333 |
+
Version enrichie qui retourne les paramètres de caméra ET les coordonnées détectées.
|
334 |
+
"""
|
335 |
+
transform = T.Resize((540, 960))
|
336 |
+
|
337 |
+
# Préparer la frame pour l'inférence
|
338 |
+
frame_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
|
339 |
+
frame_pil = Image.fromarray(frame_rgb)
|
340 |
+
frame_tensor = f.to_tensor(frame_pil).float().unsqueeze(0)
|
341 |
+
|
342 |
+
if frame_tensor.size()[-1] != 960:
|
343 |
+
frame_tensor = transform(frame_tensor)
|
344 |
+
|
345 |
+
frame_tensor = frame_tensor.to(device)
|
346 |
+
b, c, h, w = frame_tensor.size()
|
347 |
+
|
348 |
+
# Inférence
|
349 |
+
with torch.no_grad():
|
350 |
+
heatmaps = model(frame_tensor)
|
351 |
+
heatmaps_l = model_l(frame_tensor)
|
352 |
+
|
353 |
+
# Extraire les keypoints et lignes
|
354 |
+
kp_coords = get_keypoints_from_heatmap_batch_maxpool(heatmaps[:,:-1,:,:])
|
355 |
+
line_coords = get_keypoints_from_heatmap_batch_maxpool_l(heatmaps_l[:,:-1,:,:])
|
356 |
+
kp_dict = coords_to_dict(kp_coords, threshold=KP_THRESHOLD)
|
357 |
+
lines_dict = coords_to_dict(line_coords, threshold=LINE_THRESHOLD)
|
358 |
+
kp_dict, lines_dict = complete_keypoints(kp_dict[0], lines_dict[0], w=w, h=h, normalize=True)
|
359 |
+
|
360 |
+
# Calibration
|
361 |
+
cam = FramebyFrameCalib(iwidth=frame_width, iheight=frame_height, denormalize=True)
|
362 |
+
cam.update(kp_dict, lines_dict)
|
363 |
+
final_params_dict = cam.heuristic_voting(refine_lines=PNL_REFINE)
|
364 |
+
|
365 |
+
return final_params_dict, kp_dict, lines_dict
|
366 |
+
|
367 |
+
@app.get("/")
|
368 |
+
async def root():
|
369 |
+
return {
|
370 |
+
"message": "Football Vision Calibration API",
|
371 |
+
"version": "1.0.0",
|
372 |
+
"endpoints": {
|
373 |
+
"/calibrate": "POST - Calibrer une caméra à partir d'une image et de lignes",
|
374 |
+
"/inference/image": "POST - Extraire les paramètres de caméra d'une image automatiquement",
|
375 |
+
"/inference/video": "POST - Extraire les paramètres de caméra d'une vidéo automatiquement",
|
376 |
+
"/health": "GET - Vérifier l'état de l'API"
|
377 |
+
}
|
378 |
+
}
|
379 |
+
|
380 |
+
@app.get("/health")
|
381 |
+
async def health_check():
|
382 |
+
return {"status": "healthy", "message": "API is running"}
|
383 |
+
|
384 |
+
@app.post("/calibrate", response_model=CalibrationResponse)
|
385 |
+
async def calibrate_camera(
|
386 |
+
image: UploadFile = File(..., description="Image du terrain de football"),
|
387 |
+
lines_data: str = Form(..., description="JSON des lignes du terrain")
|
388 |
+
):
|
389 |
+
"""
|
390 |
+
Calibrer une caméra à partir d'une image et des lignes du terrain.
|
391 |
+
Retourne aussi les coordonnées détectées sur l'image et le terrain.
|
392 |
+
"""
|
393 |
+
try:
|
394 |
+
# Validation du format d'image - version robuste
|
395 |
+
content_type = getattr(image, 'content_type', None) or ""
|
396 |
+
filename = getattr(image, 'filename', "") or ""
|
397 |
+
|
398 |
+
# Vérifier le type MIME ou l'extension du fichier
|
399 |
+
image_extensions = ['.jpg', '.jpeg', '.png', '.bmp', '.tiff', '.tif']
|
400 |
+
is_image_content = content_type.startswith('image/') if content_type else False
|
401 |
+
is_image_extension = any(filename.lower().endswith(ext) for ext in image_extensions)
|
402 |
+
|
403 |
+
if not is_image_content and not is_image_extension:
|
404 |
+
raise HTTPException(
|
405 |
+
status_code=400,
|
406 |
+
detail=f"Le fichier doit être une image. Type détecté: {content_type}, Fichier: {filename}"
|
407 |
+
)
|
408 |
+
|
409 |
+
# Parse des données de lignes
|
410 |
+
try:
|
411 |
+
lines_dict = json.loads(lines_data)
|
412 |
+
except json.JSONDecodeError:
|
413 |
+
raise HTTPException(status_code=400, detail="Format JSON invalide pour les lignes")
|
414 |
+
|
415 |
+
# Validation de la structure des lignes
|
416 |
+
validated_lines = {}
|
417 |
+
for line_name, points in lines_dict.items():
|
418 |
+
if not isinstance(points, list):
|
419 |
+
raise HTTPException(
|
420 |
+
status_code=400,
|
421 |
+
detail=f"Les points de la ligne '{line_name}' doivent être une liste"
|
422 |
+
)
|
423 |
+
|
424 |
+
validated_points = []
|
425 |
+
for i, point in enumerate(points):
|
426 |
+
if not isinstance(point, dict) or 'x' not in point or 'y' not in point:
|
427 |
+
raise HTTPException(
|
428 |
+
status_code=400,
|
429 |
+
detail=f"Point {i} de la ligne '{line_name}' doit avoir les clés 'x' et 'y'"
|
430 |
+
)
|
431 |
+
try:
|
432 |
+
validated_points.append({
|
433 |
+
"x": float(point['x']),
|
434 |
+
"y": float(point['y'])
|
435 |
+
})
|
436 |
+
except (ValueError, TypeError):
|
437 |
+
raise HTTPException(
|
438 |
+
status_code=400,
|
439 |
+
detail=f"Coordonnées invalides pour le point {i} de la ligne '{line_name}'"
|
440 |
+
)
|
441 |
+
|
442 |
+
validated_lines[line_name] = validated_points
|
443 |
+
|
444 |
+
# Sauvegarde temporaire de l'image
|
445 |
+
file_extension = os.path.splitext(filename)[1] if filename else '.jpg'
|
446 |
+
with tempfile.NamedTemporaryFile(delete=False, suffix=file_extension) as temp_file:
|
447 |
+
content = await image.read()
|
448 |
+
temp_file.write(content)
|
449 |
+
temp_image_path = temp_file.name
|
450 |
+
|
451 |
+
try:
|
452 |
+
# Validation de l'image
|
453 |
+
pil_image = Image.open(temp_image_path)
|
454 |
+
pil_image.verify()
|
455 |
+
|
456 |
+
# Calibration de la caméra
|
457 |
+
camera_params = get_camera_parameters(temp_image_path, validated_lines)
|
458 |
+
|
459 |
+
# Pour l'endpoint calibrate, nous n'avons pas de détections automatiques
|
460 |
+
# mais nous pouvons traiter les lignes manuelles fournies
|
461 |
+
detections = None
|
462 |
+
|
463 |
+
# Formatage de la réponse
|
464 |
+
response = CalibrationResponse(
|
465 |
+
status="success",
|
466 |
+
camera_parameters=camera_params,
|
467 |
+
input_lines=validated_lines,
|
468 |
+
detections=detections,
|
469 |
+
message="Calibration réussie"
|
470 |
+
)
|
471 |
+
|
472 |
+
return response
|
473 |
+
|
474 |
+
except Exception as e:
|
475 |
+
raise HTTPException(
|
476 |
+
status_code=500,
|
477 |
+
detail=f"Erreur lors de la calibration: {str(e)}"
|
478 |
+
)
|
479 |
+
|
480 |
+
finally:
|
481 |
+
# Nettoyage du fichier temporaire
|
482 |
+
if os.path.exists(temp_image_path):
|
483 |
+
os.unlink(temp_image_path)
|
484 |
+
|
485 |
+
except HTTPException:
|
486 |
+
raise
|
487 |
+
except Exception as e:
|
488 |
+
raise HTTPException(status_code=500, detail=f"Erreur interne: {str(e)}")
|
489 |
+
|
490 |
+
@app.post("/inference/image", response_model=InferenceImageResponse)
|
491 |
+
async def inference_image(
|
492 |
+
image: UploadFile = File(..., description="Image du terrain de football"),
|
493 |
+
kp_threshold: float = Form(KP_THRESHOLD, description="Seuil pour les keypoints"),
|
494 |
+
line_threshold: float = Form(LINE_THRESHOLD, description="Seuil pour les lignes")
|
495 |
+
):
|
496 |
+
"""
|
497 |
+
Extraire automatiquement les paramètres de caméra à partir d'une image.
|
498 |
+
Retourne les coordonnées détectées sur l'image et le terrain.
|
499 |
+
"""
|
500 |
+
params = None
|
501 |
+
temp_image_path = None
|
502 |
+
|
503 |
+
try:
|
504 |
+
# Validation du format d'image - version robuste
|
505 |
+
content_type = getattr(image, 'content_type', None) or ""
|
506 |
+
filename = getattr(image, 'filename', "") or ""
|
507 |
+
|
508 |
+
# Vérifier le type MIME ou l'extension du fichier
|
509 |
+
image_extensions = ['.jpg', '.jpeg', '.png', '.bmp', '.tiff', '.tif']
|
510 |
+
is_image_content = content_type.startswith('image/') if content_type else False
|
511 |
+
is_image_extension = any(filename.lower().endswith(ext) for ext in image_extensions)
|
512 |
+
|
513 |
+
if not is_image_content and not is_image_extension:
|
514 |
+
raise HTTPException(
|
515 |
+
status_code=400,
|
516 |
+
detail=f"Le fichier doit être une image. Type détecté: {content_type}, Fichier: {filename}"
|
517 |
+
)
|
518 |
+
|
519 |
+
# Sauvegarde temporaire de l'image
|
520 |
+
file_extension = os.path.splitext(filename)[1] if filename else '.jpg'
|
521 |
+
with tempfile.NamedTemporaryFile(delete=False, suffix=file_extension) as temp_file:
|
522 |
+
content = await image.read()
|
523 |
+
temp_file.write(content)
|
524 |
+
temp_image_path = temp_file.name
|
525 |
+
|
526 |
+
# Charger les modèles
|
527 |
+
model, model_l, device = load_inference_models()
|
528 |
+
|
529 |
+
# Lire l'image
|
530 |
+
frame = cv2.imread(temp_image_path)
|
531 |
+
if frame is None:
|
532 |
+
raise HTTPException(status_code=400, detail="Impossible de lire l'image")
|
533 |
+
|
534 |
+
frame_height, frame_width = frame.shape[:2]
|
535 |
+
|
536 |
+
# Mettre à jour les seuils globaux
|
537 |
+
global KP_THRESHOLD, LINE_THRESHOLD
|
538 |
+
KP_THRESHOLD = kp_threshold
|
539 |
+
LINE_THRESHOLD = line_threshold
|
540 |
+
|
541 |
+
# Traitement avec coordonnées
|
542 |
+
params, kp_dict, lines_dict = process_frame_inference_with_coords(
|
543 |
+
frame, model, model_l, device, frame_width, frame_height
|
544 |
+
)
|
545 |
+
|
546 |
+
# Traiter les détections (coordonnées image + terrain)
|
547 |
+
detections = process_detections(kp_dict, lines_dict, params)
|
548 |
+
|
549 |
+
# Formatage de la réponse
|
550 |
+
response = InferenceImageResponse(
|
551 |
+
status="success" if params is not None else "failed",
|
552 |
+
camera_parameters=params,
|
553 |
+
image_info={
|
554 |
+
"filename": filename,
|
555 |
+
"width": frame_width,
|
556 |
+
"height": frame_height,
|
557 |
+
"kp_threshold": kp_threshold,
|
558 |
+
"line_threshold": line_threshold
|
559 |
+
},
|
560 |
+
detections=detections,
|
561 |
+
message="Paramètres extraits avec succès" if params is not None else "Échec de l'extraction des paramètres"
|
562 |
+
)
|
563 |
+
|
564 |
+
return response
|
565 |
+
|
566 |
+
except HTTPException:
|
567 |
+
raise
|
568 |
+
except Exception as e:
|
569 |
+
raise HTTPException(
|
570 |
+
status_code=500,
|
571 |
+
detail=f"Erreur lors de l'inférence: {str(e)}"
|
572 |
+
)
|
573 |
+
finally:
|
574 |
+
# Nettoyage du fichier temporaire
|
575 |
+
if temp_image_path and os.path.exists(temp_image_path):
|
576 |
+
os.unlink(temp_image_path)
|
577 |
+
|
578 |
+
@app.post("/inference/video", response_model=InferenceVideoResponse)
|
579 |
+
async def inference_video(
|
580 |
+
video: UploadFile = File(..., description="Vidéo du terrain de football"),
|
581 |
+
kp_threshold: float = Form(KP_THRESHOLD, description="Seuil pour les keypoints"),
|
582 |
+
line_threshold: float = Form(LINE_THRESHOLD, description="Seuil pour les lignes"),
|
583 |
+
frame_step: int = Form(FRAME_STEP, description="Traiter 1 frame sur N")
|
584 |
+
):
|
585 |
+
"""
|
586 |
+
Extraire automatiquement les paramètres de caméra à partir d'une vidéo.
|
587 |
+
Retourne les coordonnées détectées pour chaque frame traitée.
|
588 |
+
"""
|
589 |
+
try:
|
590 |
+
# Validation du format vidéo - version robuste
|
591 |
+
content_type = getattr(video, 'content_type', None) or ""
|
592 |
+
filename = getattr(video, 'filename', "") or ""
|
593 |
+
|
594 |
+
video_extensions = ['.mp4', '.avi', '.mov', '.mkv', '.wmv', '.flv']
|
595 |
+
is_video_content = content_type.startswith('video/') if content_type else False
|
596 |
+
is_video_extension = any(filename.lower().endswith(ext) for ext in video_extensions)
|
597 |
+
|
598 |
+
if not is_video_content and not is_video_extension:
|
599 |
+
raise HTTPException(
|
600 |
+
status_code=400,
|
601 |
+
detail=f"Le fichier doit être une vidéo. Type détecté: {content_type}, Fichier: {filename}"
|
602 |
+
)
|
603 |
+
|
604 |
+
# Sauvegarde temporaire de la vidéo
|
605 |
+
file_extension = os.path.splitext(filename)[1] if filename else '.mp4'
|
606 |
+
with tempfile.NamedTemporaryFile(delete=False, suffix=file_extension) as temp_file:
|
607 |
+
content = await video.read()
|
608 |
+
temp_file.write(content)
|
609 |
+
temp_video_path = temp_file.name
|
610 |
+
|
611 |
+
try:
|
612 |
+
# Charger les modèles
|
613 |
+
model, model_l, device = load_inference_models()
|
614 |
+
|
615 |
+
# Ouvrir la vidéo
|
616 |
+
cap = cv2.VideoCapture(temp_video_path)
|
617 |
+
if not cap.isOpened():
|
618 |
+
raise HTTPException(status_code=400, detail="Impossible d'ouvrir la vidéo")
|
619 |
+
|
620 |
+
frame_width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
|
621 |
+
frame_height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
|
622 |
+
total_frames = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
|
623 |
+
fps = int(cap.get(cv2.CAP_PROP_FPS))
|
624 |
+
|
625 |
+
# Mettre à jour les seuils globaux
|
626 |
+
global KP_THRESHOLD, LINE_THRESHOLD
|
627 |
+
KP_THRESHOLD = kp_threshold
|
628 |
+
LINE_THRESHOLD = line_threshold
|
629 |
+
|
630 |
+
frames_results = []
|
631 |
+
frame_count = 0
|
632 |
+
processed_count = 0
|
633 |
+
|
634 |
+
while cap.isOpened():
|
635 |
+
ret, frame = cap.read()
|
636 |
+
if not ret:
|
637 |
+
break
|
638 |
+
|
639 |
+
# Traiter seulement 1 frame sur frame_step
|
640 |
+
if frame_count % frame_step != 0:
|
641 |
+
frame_count += 1
|
642 |
+
continue
|
643 |
+
|
644 |
+
# Traitement avec coordonnées
|
645 |
+
params, kp_dict, lines_dict = process_frame_inference_with_coords(
|
646 |
+
frame, model, model_l, device, frame_width, frame_height
|
647 |
+
)
|
648 |
+
|
649 |
+
if params is not None:
|
650 |
+
# Traiter les détections (coordonnées image + terrain)
|
651 |
+
detections = process_detections(kp_dict, lines_dict, params)
|
652 |
+
|
653 |
+
frame_result = FrameResult(
|
654 |
+
frame_number=frame_count,
|
655 |
+
timestamp_seconds=frame_count / fps,
|
656 |
+
camera_parameters=params,
|
657 |
+
detections=detections
|
658 |
+
)
|
659 |
+
|
660 |
+
frames_results.append(frame_result)
|
661 |
+
processed_count += 1
|
662 |
+
|
663 |
+
frame_count += 1
|
664 |
+
|
665 |
+
cap.release()
|
666 |
+
|
667 |
+
# Formatage de la réponse
|
668 |
+
response = InferenceVideoResponse(
|
669 |
+
status="success" if frames_results else "failed",
|
670 |
+
video_info={
|
671 |
+
"filename": filename,
|
672 |
+
"width": frame_width,
|
673 |
+
"height": frame_height,
|
674 |
+
"total_frames": total_frames,
|
675 |
+
"fps": fps,
|
676 |
+
"duration_seconds": total_frames / fps,
|
677 |
+
"kp_threshold": kp_threshold,
|
678 |
+
"line_threshold": line_threshold,
|
679 |
+
"frame_step": frame_step
|
680 |
+
},
|
681 |
+
frames_processed=processed_count,
|
682 |
+
frames_results=frames_results,
|
683 |
+
message=f"Paramètres extraits de {processed_count} frames" if frames_results else "Aucun paramètre extrait"
|
684 |
+
)
|
685 |
+
|
686 |
+
return response
|
687 |
+
|
688 |
+
except Exception as e:
|
689 |
+
raise HTTPException(
|
690 |
+
status_code=500,
|
691 |
+
detail=f"Erreur lors de l'inférence vidéo: {str(e)}"
|
692 |
+
)
|
693 |
+
|
694 |
+
finally:
|
695 |
+
# Nettoyage du fichier temporaire
|
696 |
+
if os.path.exists(temp_video_path):
|
697 |
+
os.unlink(temp_video_path)
|
698 |
+
|
699 |
+
except HTTPException:
|
700 |
+
raise
|
701 |
+
except Exception as e:
|
702 |
+
raise HTTPException(status_code=500, detail=f"Erreur interne: {str(e)}")
|
703 |
+
|
704 |
app_instance = app
|
requirements.txt
CHANGED
@@ -1,25 +1,23 @@
|
|
1 |
-
# API Framework
|
2 |
-
fastapi==0.104.1
|
3 |
-
uvicorn[standard]==0.24.0
|
4 |
-
python-multipart==0.0.6
|
5 |
-
pydantic==2.5.0
|
6 |
-
|
7 |
-
# Core dependencies (compatibles Python 3.10)
|
8 |
-
numpy==1.24.3
|
9 |
-
opencv-python-headless==4.8.1.78
|
10 |
-
pillow==10.1.0
|
11 |
-
scipy==1.11.4
|
12 |
-
PyYAML==6.0.1
|
13 |
-
lsq-ellipse==2.2.1
|
14 |
-
shapely==2.0.2
|
15 |
-
|
16 |
-
# PyTorch CPU (compatible Python 3.10)
|
17 |
-
torch==2.1.0
|
18 |
-
torchvision==0.16.0
|
19 |
-
|
20 |
-
# Utilities
|
21 |
-
tqdm==4.66.1
|
22 |
-
|
23 |
-
|
24 |
-
|
25 |
huggingface_hub
|
|
|
1 |
+
# API Framework
|
2 |
+
fastapi==0.104.1
|
3 |
+
uvicorn[standard]==0.24.0
|
4 |
+
python-multipart==0.0.6
|
5 |
+
pydantic==2.5.0
|
6 |
+
|
7 |
+
# Core dependencies (compatibles Python 3.10)
|
8 |
+
numpy==1.24.3
|
9 |
+
opencv-python-headless==4.8.1.78
|
10 |
+
pillow==10.1.0
|
11 |
+
scipy==1.11.4
|
12 |
+
PyYAML==6.0.1
|
13 |
+
lsq-ellipse==2.2.1
|
14 |
+
shapely==2.0.2
|
15 |
+
|
16 |
+
# PyTorch CPU (compatible Python 3.10)
|
17 |
+
torch==2.1.0
|
18 |
+
torchvision==0.16.0
|
19 |
+
|
20 |
+
# Utilities
|
21 |
+
tqdm==4.66.1
|
22 |
+
|
|
|
|
|
23 |
huggingface_hub
|