Inference speed issue

[bgyooPtr]

Search before asking

• I have searched the YOLOv5 issues and found no similar bug report.

YOLOv5 Component

Detection, Integrations

Bug

When executing real-time prediction using Realsense d415, there is a big difference in speed depending on the camera fps.
The higher the camera FPS, the slower the inference speed.
Obviously, I wonder why this happens even though the function that reads data is composed of threads.

• fps: 30
  avg time: 10.963139772415161
• fps: 60
  avg time: 24.456851720809937
• fps: 90
  avg time: 35.522178649902344

Environment

• detect: weights=['runs/train/exp2/weights/best.pt'], data=data/coco128.yaml, imgsz=[500, 500], conf_thres=0.25, iou_thres=0.45, max_det=1000, device=, view_img=False, save_txt=False, save_conf=False, save_crop=False, nosave=False, classes=None, agnostic_nms=False, augment=False, visualize=False, update=False, project=runs/detect, name=exp, exist_ok=False, line_thickness=3, hide_labels=False, hide_conf=False, half=False, dnn=False, vid_stride=1
• YOLOv5 🚀 v6.2-211-g32a9218 Python-3.7.13 torch-1.8.1 CUDA:0 (NVIDIA TITAN RTX, 24217MiB)

Fusing layers...
Model summary: 157 layers, 1760518 parameters, 0 gradients, 4.1 GFLOPs
WARNING ⚠️ --img-size [500, 500] must be multiple of max stride 32, updating to [512, 512]

Minimal Reproducible Example

class LoadStreams1:
    # YOLOv5 streamloader, i.e. `python detect.py --source 'rtsp://example.com/media.mp4'  # RTSP, RTMP, HTTP streams`
    def __init__(
        self,
        img_size=640,
        stride=32,
        auto=True,
        transforms=None,
        vid_stride=1,
        fps=30,
    ):
        torch.backends.cudnn.benchmark = True  # faster for fixed-size inference
        self.img_size = img_size
        self.stride = stride
        self.vid_stride = vid_stride  # video frame-rate stride
        self.img, self.depth, self.fps, self.frame, self.thread = (
            None,
            None,
            0,
            0,
            None,
        )

        self.pipeline = rs.pipeline()
        config = rs.config()
        config.enable_stream(rs.stream.color, 848, 480, rs.format.rgb8, fps)
        config.enable_stream(rs.stream.depth, 848, 480, rs.format.z16, fps)
        self.align_to_color = rs.align(rs.stream.color)

        self.prof = self.pipeline.start(config)
        device = self.prof.get_device().first_depth_sensor()
        preset_range = device.get_option_range(rs.option.visual_preset)
        for i in range(int(preset_range.max)):
            visulpreset = device.get_option_value_description(
                rs.option.visual_preset, i
            )
            print("%02d: %s" % (i, visulpreset))
            if visulpreset == "High Accuracy":
                device.set_option(rs.option.visual_preset, i)
                print(":: set preset to High Accuracy")

        # warm up
        for i in range(60):
            # ret, frames = self.pipeline.wait_for_frames()
            ret, frames = self.pipeline.try_wait_for_frames()
            if not ret:
                print("warm up failed")
                continue
            aligned_frames = self.align_to_color.process(frames)
            color_frame = aligned_frames.get_color_frame()
            rgb_image = np.asanyarray(color_frame.get_data())
            bgr_image = cv2.cvtColor(rgb_image, cv2.COLOR_RGB2BGR)
            self.img = bgr_image

            depth_frame = aligned_frames.get_depth_frame()
            depth_data = np.array(depth_frame.data)
            self.depth = depth_data

            break

        self.thread = threading.Thread(
            target=self.update, args=(), daemon=True
        )
        self.thread.start()
        while True:
            if self.thread.is_alive():
                break
            time.sleep(0.1)

        # check for common shapes
        s = np.stack(
            [
                letterbox(self.img, img_size, stride=stride, auto=auto)[0].shape
            ]
        )
        self.rect = (
            np.unique(s, axis=0).shape[0] == 1
        )  # rect inference if all shapes equal
        self.auto = auto and self.rect
        self.transforms = transforms  # optional
        if not self.rect:
            LOGGER.warning(
                "WARNING ⚠️ Stream shapes differ. For optimal performance supply similarly-shaped streams."
            )

    def update(self):
        while True:
            ret, frames = self.pipeline.try_wait_for_frames()
            if not ret:
                LOGGER.warning(
                    "WARNING ⚠️ Video stream unresponsive, please check your IP camera connection."
                )
                break
            aligned_frames = self.align_to_color.process(frames)
            color_frame = aligned_frames.get_color_frame()
            rgb_image = np.asanyarray(color_frame.get_data())
            bgr_image = cv2.cvtColor(rgb_image, cv2.COLOR_RGB2BGR)

            depth_frame = aligned_frames.get_depth_frame()
            depth_data = np.array(depth_frame.data)

            self.imgs = bgr_image
            self.depth = depth_data
            time.sleep(0.0)  # wait time

    def __iter__(self):
        self.count = -1
        return self

    def __next__(self):
        self.count += 1
        if not self.thread.is_alive() or cv2.waitKey(1) == ord(
            "q"
        ):  # q to quit
            cv2.destroyAllWindows()
            raise StopIteration

        im0 = self.img.copy()
        depthu = None
        if isinstance(self.depth, np.ndarray):
            depthu = self.depth.copy()
        im = np.stack(
            [
                letterbox(im0, self.img_size, stride=self.stride, auto=self.auto)[0]
            ]
        )  # resize
        im = im[..., ::-1].transpose((0, 3, 1, 2))  # BGR to RGB, BHWC to BCHW
        im = np.ascontiguousarray(im)  # contiguous

        return im, im0, depthu

    # Load model
    device = select_device(device)
    model = DetectMultiBackend(weights, device=device, dnn=dnn, data=data, fp16=half)
    stride, names, pt = model.stride, model.names, model.pt
    imgsz = check_img_size(imgsz, s=stride)  # check image size

    # Dataloader
    bs = 1  # batch_size

    # Run inference
    model.warmup(imgsz=(1 if pt or model.triton else bs, 3, *imgsz))  # warmup
    dataset = LoadStreams1(
         img_size=imgsz, stride=stride, auto=pt, vid_stride=vid_stride, fps=90
    )
    exit(0)
    times = []
    for  _, im0s, depth in dataset:
        if len(times) == 1000:
            break
        s_time = time.time()
        img0 = letterbox(im0s, imgsz, stride, pt)[0]
        img = img0.transpose((2, 0, 1))[::-1]  # HWC to CHW, BGR to RGB
        img = np.ascontiguousarray(img)
        im = torch.from_numpy(img).to(model.device)
        im = im.half() if model.fp16 else im.float()  # uint8 to fp16/32
        im /= 255  # 0 - 255 to 0.0 - 1.0
        if len(im.shape) == 3:
            im = im[None]  # expand for batch dim

        # Inference
        pred = model(im, augment=augment, visualize=False)

        # NMS
        pred = non_max_suppression(
            pred, conf_thres, iou_thres, classes, agnostic_nms, max_det=max_det
        )

        e_time = (time.time() - s_time) * 1000
        times.append(e_time)
        print("elapsed time: ", e_time)
    print("avg time: ", np.mean(times))

Additional

No response

Are you willing to submit a PR?

• Yes I'd like to help by submitting a PR!

[github-actions]

👋 Hello @bgyooPtr, thank you for your interest in YOLOv5 🚀! Please visit our ⭐️ Tutorials to get started, where you can find quickstart guides for simple tasks like Custom Data Training all the way to advanced concepts like Hyperparameter Evolution.

If this is a 🐛 Bug Report, please provide screenshots and minimum viable code to reproduce your issue, otherwise we can not help you.

If this is a custom training ❓ Question, please provide as much information as possible, including dataset images, training logs, screenshots, and a public link to online W&B logging if available.

For business inquiries or professional support requests please visit https://ultralytics.com or email support@ultralytics.com.

Requirements

Python>=3.7.0 with all requirements.txt installed including PyTorch>=1.7. To get started:

git clone https://github.com/ultralytics/yolov5  # clone
cd yolov5
pip install -r requirements.txt  # install

Environments

YOLOv5 may be run in any of the following up-to-date verified environments (with all dependencies including CUDA/CUDNN, Python and PyTorch preinstalled):

• Notebooks with free GPU:
• Google Cloud Deep Learning VM. See GCP Quickstart Guide
• Amazon Deep Learning AMI. See AWS Quickstart Guide
• Docker Image. See Docker Quickstart Guide

Status

If this badge is green, all YOLOv5 GitHub Actions Continuous Integration (CI) tests are currently passing. CI tests verify correct operation of YOLOv5 training, validation, inference, export and benchmarks on MacOS, Windows, and Ubuntu every 24 hours and on every commit.

[glenn-jocher]

👋 Hello! Thanks for asking about inference speed issues. PyTorch Hub speeds will vary by hardware, software, model, inference settings, etc. Our default example in Colab with a V100 looks like this:

YOLOv5 🚀 can be run on CPU (i.e. --device cpu, slow) or GPU if available (i.e. --device 0, faster). You can determine your inference device by viewing the YOLOv5 console output:

detect.py inference

python detect.py --weights yolov5s.pt --img 640 --conf 0.25 --source data/images/

YOLOv5 PyTorch Hub inference

import torch

# Model
model = torch.hub.load('ultralytics/yolov5', 'yolov5s')

# Images
dir = 'https://ultralytics.com/images/'
imgs = [dir + f for f in ('zidane.jpg', 'bus.jpg')]  # batch of images

# Inference
results = model(imgs)
results.print()  # or .show(), .save()
# Speed: 631.5ms pre-process, 19.2ms inference, 1.6ms NMS per image at shape (2, 3, 640, 640)

Increase Speeds

If you would like to increase your inference speed some options are:

• Use batched inference with YOLOv5 PyTorch Hub
• Reduce --img-size, i.e. 1280 -> 640 -> 320
• Reduce model size, i.e. YOLOv5x -> YOLOv5l -> YOLOv5m -> YOLOv5s -> YOLOv5n
• Use half precision FP16 inference with python detect.py --half and python val.py --half
• Use a faster GPUs, i.e.: P100 -> V100 -> A100
• Export to ONNX or OpenVINO for up to 3x CPU speedup (CPU Benchmarks)
• Export to TensorRT for up to 5x GPU speedup (GPU Benchmarks)
• Use a free GPU backends with up to 16GB of CUDA memory:

Good luck 🍀 and let us know if you have any other questions!

[github-actions]

👋 Hello, this issue has been automatically marked as stale because it has not had recent activity. Please note it will be closed if no further activity occurs.

Access additional YOLOv5 🚀 resources:

• Wiki – https://github.com/ultralytics/yolov5/wiki
• Tutorials – https://docs.ultralytics.com/yolov5
• Docs – https://docs.ultralytics.com

Access additional Ultralytics ⚡ resources:

• Ultralytics HUB – https://ultralytics.com/hub
• Vision API – https://ultralytics.com/yolov5
• About Us – https://ultralytics.com/about
• Join Our Team – https://ultralytics.com/work
• Contact Us – https://ultralytics.com/contact

Feel free to inform us of any other issues you discover or feature requests that come to mind in the future. Pull Requests (PRs) are also always welcomed!

Thank you for your contributions to YOLOv5 🚀 and Vision AI ⭐!