YOLO-POSE code

[chophilip21]

❔Question

Hi, I have read your paper: "YOLO-Pose: Enhancing YOLO for Multi Person Pose Estimation Using Object
Keypoint Similarity Loss", and I was fascinated by the potential of your idea. Your paper is pointing at this repo, but I cannot seem to find any material related to keypoint detection. All I see is your code related to object detection.

Are you still working on this? Or does the code exist in another branch?

[debapriyamaji]

Hi @chophilip21 ,
Thanks for your interest. I will upload the code sometime next month. Will keep you posted on the same.
Regards, Debapriya

[chophilip21]

@debapriyamaji Thank you for your response! Will be looking forward to the code update.

[Bestsongc]

Hi, I have read your paper: "YOLO-Pose: Enhancing YOLO for Multi Person Pose Estimation Using Object
Keypoint Similarity Loss", and I was fascinated by the potential of your idea. i am looking forward to the code update

[jamjamjon]

When will release pose code？ Anticipating it!

[wangduyang]

nice work！
waitting for your code！

[sly985-love]

waitting for your code!

[matveymor]

nice work! waiting for your code and pretrained weights 🙏

[a-dubbel]

+1! 🙏

[backermanaaa]

Hello, when will release pose code？ Anticipating it!

[debapriyamaji]

Thanks, everyone for your interest. I will release the code by end of this month.

Regards, Debapriya

[yizhangliu]

Thanks for your work. Waitting for your code!

[wijjj]

Would be VERY curious as well! Thanks in advance.

[BaofengZan]

Waitting for your code!

[ZhenpengChenCode]

Thanks for your work. Waiting for your code!

[paleomoon]

Thanks for your work. Waiting for your code!

[xiakj]

watching!

[TanateT]

waiting for u

[yizhangliu]

Maybe we can find what we need from https://github.com/TexasInstruments/edgeai-yolox:
human pose and keypoints.

[jamjamjon]

Maybe we can find what we need from https://github.com/TexasInstruments/edgeai-yolox:
human pose and keypoints.
Nothing there！ ？

[jamjamjon]

It’s the end of May, time to release the code! Appreciate it.

[debapriyamaji]

Hi All,
Thanks for your interest. I am working on it, I am trying to release it by this weekend. Sorry for the little delay.

Regards, Debapriya

[Dramazy]

keep waiting!

[debapriyamaji]

Hi All,

Please try out the yolo-pose code from here:
https://github.com/TexasInstruments/edgeai-yolov5/tree/yolo-pose

If you find this repo useful, please consider giving it a star.

Regards, Debapriya

[yizhangliu]

You are prefect guy. Thanks.

[bui-thanh-lam]

Are there any pretrained weights released? I think all the checkpoint URL from Readme are disabled.

[dizcza]

All links are dead.

Has anyone downloaded the yolov5 pose weights? Please share.

Thanks.

[matveymorozov]

@dizcza

It appears that there may be an issue with the links in question. If you encounter an error when clicking on a link, please try copying the link URL directly from the name of the section in which it is located. Specifically, the correct labels can be found within the YOLOv5-ti-lite Based Models and Ckpts section.

If copying the URL from the section name does not work, you may need to manually enter the URL into your web browser's search bar.

Small 640:
http://software-dl.ti.com/jacinto7/esd/modelzoo/gplv3/08_02_00_11/edgeai-yolov5/pretrained_models/checkpoints/keypoint/coco/edgeai-yolov5/yolov5s6_640_ti_lite_54p9_82p2/weights/last.pt

Small 960:
http://software-dl.ti.com/jacinto7/esd/modelzoo/gplv3/08_02_00_11/edgeai-yolov5/pretrained_models/checkpoints/keypoint/coco/edgeai-yolov5/yolov5s6_960_ti_lite_59p7_85p6/weights/last.pt

Medium 640:
http://software-dl.ti.com/jacinto7/esd/modelzoo/gplv3/08_02_00_11/edgeai-yolov5/pretrained_models/checkpoints/keypoint/coco/edgeai-yolov5/yolov5m6_640_ti_lite_60p5_86p8/weights/best.pt

Medium 960:
http://software-dl.ti.com/jacinto7/esd/modelzoo/gplv3/08_02_00_11/edgeai-yolov5/pretrained_models/checkpoints/keypoint/coco/edgeai-yolov5/yolov5m6_960_ti_lite_65p9_88p6/weights/last.pt

[dizcza]

@matveymorozov thanks for the hint.

My Chrome browser still refuses to download the links. But I managed to do the trick with wget.

[matveymorozov]

@dizcza try to copy it and manually pass in search bar

[dizcza]

try to copy it and manually pass in search bar

That's exactly what I did. It doesn't work for me.

Anyhow, I successfully downloaded the file with wget.

[dizcza]

@matveymorozov is it possible to run yolov5s-pose in OpenCV DNN rather than onnxruntime? I've downloaded ONNX-converted model of yours:

    model_path = "yolov5s6_pose_640_ti_lite_54p9_82p2.onnx"
    net = cv2.dnn.readNet(model_path)

  File "/home/dizcza/Projects/Airtouch/edgeai-yolov5/onnx_inference/yolo_pose_onnx_inference.py", line 143, in inference_video
    net = cv2.dnn.readNet(model_path)
cv2.error: OpenCV(4.7.0) /io/opencv/modules/dnn/src/onnx/onnx_importer.cpp:1073: error: (-2:Unspecified error) in function 'handleNode'
> Node [Split@ai.onnx]:(onnx_node!Split_1904) parse error: OpenCV(4.7.0) /io/opencv/modules/dnn/src/onnx/onnx_importer.cpp:1468: error: (-215:Assertion failed) numSplits > 1 in function 'parseSplit'

I also tried converting the same model to ONNX manually and got:

$ python models/export.py --weights last.pt  --img 640 --batch 1 --simplify --export-nms
Namespace(weights='last.pt', img_size=[640, 640], batch_size=1, grid=False, device='cpu', dynamic=False, simplify=True, export_nms=True)
YOLOv5 � v4.0-76-gae4e0e8 torch 2.0.0+cu117 CPU

Fusing layers... 
Model Summary: 436 layers, 15086540 parameters, 0 gradients
Traceback (most recent call last):
  File "/home/dizcza/Projects/Airtouch/edgeai-yolov5/models/export.py", line 71, in <module>
    y = model(img)  # dry runs
  File "/home/dizcza/miniconda3/envs/yolov7-pose/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl
    return forward_call(*args, **kwargs)
  File "/home/dizcza/Projects/Airtouch/edgeai-yolov5/models/yolo.py", line 157, in forward
    return self.forward_once(x, profile)  # single-scale inference, train
  File "/home/dizcza/Projects/Airtouch/edgeai-yolov5/models/yolo.py", line 188, in forward_once
    x = m(x)  # run
  File "/home/dizcza/miniconda3/envs/yolov7-pose/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl
    return forward_call(*args, **kwargs)
  File "/home/dizcza/miniconda3/envs/yolov7-pose/lib/python3.10/site-packages/torch/nn/modules/upsampling.py", line 157, in forward
    recompute_scale_factor=self.recompute_scale_factor)
  File "/home/dizcza/miniconda3/envs/yolov7-pose/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1614, in __getattr__
    raise AttributeError("'{}' object has no attribute '{}'".format(
AttributeError: 'Upsample' object has no attribute 'recompute_scale_factor'

[dizcza]

Note that onnxruntime works fine. It's just I'm looking for something that can be easily used in OpenCV without any additional engines.

[matveymorozov]

Actually, I don't know. I just tested it with onnxruntime and Pytorch

[Mugutech62]

Thanks, everyone for your interest. I will release the code by end of this month.

Regards, Debapriya

I am having doubt how can i convert my yolo v5 model into yolo ti model

[MR-STUZHANG]

谢谢大家的关注。我将在本月底之前发布代码。

问候，德巴普里亚
I only saw the AP of the detection box in the code, and did not see the AP related calculation of OKS. Can you provide some technical guidance

[AliasChenYi]

I have reproduced the result of yolov8s-pose, and the experimental data is very different from yours, with map50 dropping 20% points. My code and data are exactly the same as what you described, and the operation steps are also strictly followed by yours. Therefore, I may have a problem with hyperparameter setting.
Here are my hyperparameter Settings：
task: pose
mode: train
model: ultralytics/cfg/models/v8pose/yolov8-pose.yaml
data: ultralytics/cfg/datasets/coco-pose.yaml
epochs: 300
time: null
patience: 50
batch: 310
imgsz: 640
save: true
save_period: -1
cache: true
device: '0'
workers: 20
project: runs/train
name: yolov8s-pose
exist_ok: false
pretrained: true
optimizer: SGD
verbose: false
seed: 0
deterministic: true
single_cls: true
rect: false
cos_lr: false
close_mosaic: 10
resume: false
amp: true
fraction: 1.0
profile: false
freeze: null
multi_scale: false
overlap_mask: true
mask_ratio: 4
dropout: 0.0
val: true
split: val
save_json: false
save_hybrid: false
conf: null
iou: 0.7
max_det: 300
half: false
dnn: false
plots: true
source: null
vid_stride: 1
stream_buffer: false
visualize: false
augment: false
agnostic_nms: false
classes: null
retina_masks: false
embed: null
show: false
save_frames: false
save_txt: false
save_conf: false
save_crop: false
show_labels: true
show_conf: true
show_boxes: true
line_width: null
format: torchscript
keras: false
optimize: false
int8: false
dynamic: false
simplify: false
opset: null
workspace: 4
nms: false
lr0: 0.01
lrf: 0.01
momentum: 0.937
weight_decay: 0.0005
warmup_epochs: 3.0
warmup_momentum: 0.8
warmup_bias_lr: 0.1
box: 7.5
cls: 0.5
dfl: 1.5
pose: 12.0
kobj: 1.0
label_smoothing: 0.0
nbs: 64
hsv_h: 0.015
hsv_s: 0.7
hsv_v: 0.4
degrees: 0.0
translate: 0.1
scale: 0.5
shear: 0.0
perspective: 0.0
flipud: 0.0
fliplr: 0.5
mosaic: 0.5
mixup: 0.2
copy_paste: 0.0
auto_augment: randaugment
erasing: 0.4
crop_fraction: 1.0
cfg: null
tracker: botsort.yaml
save_dir: runs/train/yolov8s-pose