How can I use deepsparse instead of ultralytics
piaoyaoi opened this issue · 2 comments
`def process_frame(frame, contours, pts,
object_code, model, class_thresh: int = 0.25,
_algorithm_queue: multiprocessing.Queue = None):
results = model.predict(source=frame, conf=class_thresh,
vid_stride=True, classes=object_code, device=device) # 对当前帧进行目标检测并显示结果
cpu_res = results[0].cpu()
boxes = cpu_res.boxes # Boxes object for bbox outputs
cls_list = boxes.cls.tolist()
# 类名字典
names_dic = cpu_res.names
have_algorithm = False
# 循环遍历所有的检测结果
for i in range(len(cls_list)):
x1, y1, x2, y2 = map(int, boxes.xyxy[i]) # 转换为整数坐标
# 边界框的四个顶点
box_points = np.array([
[x1, y1],
[x2, y1],
[x2, y2],
[x1, y2]
], dtype=np.int32)
# 使用pointPolygonTest检查至少一个边界框的顶点是否在区域内
is_inside = False
color = (0, 0, 255) # 默认红色
for box_point in box_points:
# 直接从box_point数组中提取坐标，确保它们作为单独的元素传入
# is_inside = cv2.pointPolygonTest(pts, (box_point[0], box_point[1]), False) >= 0
# 显式地将box_point的坐标转换为整数元组
point_tuple = tuple(map(int, box_point.tolist()))
is_inside = cv2.pointPolygonTest(pts, point_tuple, False) >= 0
if is_inside:
break
if is_inside:
have_algorithm = True
else:
pass
if not is_inside:
color = (0, 255, 0) # 绿色
# continue
# 获取类别名称
cls_name = names_dic[cls_list[i]]
cv2.rectangle(frame, (x1, y1), (x2, y2), color, 2)
conf = float(boxes.conf[i])
# 添加文字标签（类别+置信度）
label = f"{cls_name}: {conf:.2f}"
cv2.putText(frame, label, (x1, y1 - 10), cv2.FONT_HERSHEY_SIMPLEX, 0.5, color, 2)
cv2.polylines(frame, contours, isClosed=True, color=(0, 255, 0), thickness=2)
# 报警视频帧存入报警队列
if have_algorithm:
try:
_algorithm_queue.put_nowait(frame)
except queue.Full:
logging.log(logging.ERROR, "Algorithm queue is full")
pass
return frame
`
This is the code for yolov8 that I used earlier，Now I'm going to use deepsparse instead
`def process_frame(frame, contours, pts,
object_code, model, yolo_pipeline, class_thresh: int = 0.25,
_algorithm_queue: multiprocessing.Queue = None):
print('开始检测')
# input_tensor = preprocess_frame_for_deepsparse(frame)
# pipeline_outputs = compiled_model.run([input_tensor])
# save_pipeline_outputs_to_file(pipeline_outputs)
# print('pipeline_outputs',pipeline_outputs)
start_time = time.time()
pipeline_outputs_low_conf = yolo_pipeline(images=frame, iou_thres=0.3, conf_thres=0.1)
end_time = time.time()
execution_time = end_time - start_time
print("执行时间:", execution_time*1000, "ms")
boxes = pipeline_outputs_low_conf.boxes
scores = pipeline_outputs_low_conf.scores
labels = pipeline_outputs_low_conf.labels
print('boxes', boxes)
print('scores', scores)
print('labels', labels)`
What should I write
Hi @piaoyaoi
The code looks fine overall and includes the DeepSparse code....
Was there a bug you had encountered or something further you are trying to accomplish?
Best,
Jeannie / Neural Magic