Spaces:
Runtime error
Runtime error
File size: 5,464 Bytes
8fb6272 b181644 8fb6272 b181644 8fb6272 b181644 8fb6272 b181644 577120c b181644 8fb6272 b181644 8fb6272 b181644 8fb6272 577120c b181644 8fb6272 b181644 8fb6272 b181644 577120c 8fb6272 b181644 8fb6272 b181644 8fb6272 b181644 8fb6272 b181644 8fb6272 b181644 8fb6272 b181644 577120c b181644 8fb6272 b181644 8fb6272 b181644 8fb6272 b181644 8fb6272 b181644 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 |
import torch
import torch.nn.functional as F
import logging
import os
import gc
import numpy as np
import cv2
from PIL import Image
from transformers import Owlv2Processor, Owlv2ForObjectDetection
from .base import BaseDetector
logger = logging.getLogger(__name__)
class WeaponDetectorGPU(BaseDetector):
"""Detector de armas otimizado para GPU."""
def __init__(self):
"""Inicializa o detector."""
super().__init__()
self.default_resolution = 640
self.device = None # Será configurado em _initialize
self._initialize()
def _initialize(self):
"""Inicializa o modelo."""
try:
# Configurar device
if not torch.cuda.is_available():
raise RuntimeError("CUDA não está disponível!")
# Configurar device corretamente
self.device = 0 # Usar índice inteiro para GPU
# Carregar modelo e processador
logger.info("Carregando modelo e processador...")
model_name = "google/owlv2-base-patch16"
self.owlv2_processor = Owlv2Processor.from_pretrained(model_name)
self.owlv2_model = Owlv2ForObjectDetection.from_pretrained(
model_name,
torch_dtype=torch.float16,
device_map={"": self.device} # Mapear todo o modelo para GPU 0
)
# Otimizar modelo
self.owlv2_model.eval()
# Processar queries
self.text_queries = self._get_detection_queries()
self.processed_text = self.owlv2_processor(
text=self.text_queries,
return_tensors="pt",
padding=True
)
self.processed_text = {
key: val.to(self.device)
for key, val in self.processed_text.items()
}
logger.info("Inicialização GPU completa!")
self._initialized = True
except Exception as e:
logger.error(f"Erro na inicialização GPU: {str(e)}")
raise
def detect_objects(self, image: Image.Image, threshold: float = 0.3) -> list:
"""Detecta objetos em uma imagem."""
try:
# Pré-processar imagem
if image.mode != 'RGB':
image = image.convert('RGB')
# Processar imagem
image_inputs = self.owlv2_processor(
images=image,
return_tensors="pt"
)
image_inputs = {
key: val.to(self.device)
for key, val in image_inputs.items()
}
# Inferência
with torch.no_grad():
inputs = {**image_inputs, **self.processed_text}
outputs = self.owlv2_model(**inputs)
target_sizes = torch.tensor([image.size[::-1]], device=self.device)
results = self.owlv2_processor.post_process_grounded_object_detection(
outputs=outputs,
target_sizes=target_sizes,
threshold=threshold
)[0]
# Processar detecções
detections = []
if len(results["scores"]) > 0:
scores = results["scores"]
boxes = results["boxes"]
labels = results["labels"]
for score, box, label in zip(scores, boxes, labels):
if score.item() >= threshold:
detections.append({
"confidence": score.item(),
"box": [int(x) for x in box.tolist()],
"label": self.text_queries[label]
})
return detections
except Exception as e:
logger.error(f"Erro em detect_objects: {str(e)}")
return []
def _get_best_device(self):
"""Retorna o melhor dispositivo disponível."""
return 0 # Usar índice inteiro para GPU
def _clear_gpu_memory(self):
"""Limpa memória GPU."""
torch.cuda.empty_cache()
gc.collect()
def process_video(self, video_path: str, fps: int = None, threshold: float = 0.3, resolution: int = 640) -> tuple:
"""Processa um vídeo."""
metrics = {
"total_time": 0,
"frames_analyzed": 0,
"detections": []
}
try:
frames = self.extract_frames(video_path, fps or 2, resolution)
metrics["frames_analyzed"] = len(frames)
for i, frame in enumerate(frames):
frame_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
frame_pil = Image.fromarray(frame_rgb)
detections = self.detect_objects(frame_pil, threshold)
if detections:
metrics["detections"].append({
"frame": i,
"detections": detections
})
return video_path, metrics
return video_path, metrics
except Exception as e:
logger.error(f"Erro ao processar vídeo: {str(e)}")
return video_path, metrics |