Excuse me, How is the configuration file of train.py file generated. I want to train lspet dataset

Hi @vietanhdev, thanks for your great work! I notice that you will support the Yoga-82 dataset in the future according to your TODO list. However, when I searched the Yogo-82 dataset, I found this dataset only contains the category labels without any pose landmark annotations. Do you have the pose landmark annotations of Yoga-82 dataset?

How could I start training with my own dataset?

Hi,
Thanks for the amazing work. Quick question : is there a way to determine the optimal heatmap shape given an input image ? My images are (512, 1024, 3) and have between 3 to 7 keypoints, what heatmap size should I use ?

Cheers

Hi, great work

When I see your codes, Bbox is only needed for training but not testing.

is Bbox mandatory to use in training phase?? or can I just get rid of it to train the model?

Your model is awesome. Would you add the reference of the pushup dataset?

Hi author,

When I convert the model to onnx format, an error similar with following rises up:
ValueError: Graph disconnected: cannot obtain value for tensor Tensor("joints/conv2d_58_input:0", shape=(None, 2, 2, 288), dtype=float32) at layer "conv2d_58". The following previous layers were accessed without issue: []

Both tensorflow 2.3.0 and 2.5.0 have the same error. I did not find out ant possible reasons for this issue. Any suggestions? Thanks.

/home/sa/workspace/temp/tf-blazepose-master/src/metrics/pck.py:46 update_state
if len(tf.shape(y_true)) == 4: # Heatmap
/home/sa/anaconda3/lib/python3.6/site-packages/tensorflow/python/framework/ops.py:754 len
"shape information.".format(self.name))

TypeError: len is not well defined for symbolic Tensors.

Hi there,
I am trying to run the split_lsp_lspet.py file, seems it's not able to locate the other files referenced. src module not found.
I tried even deleting the init.py file, still the same error.

If you have the pre-trained weights for this model trained on mpii datset to share would be nice to have them for testing use. Thanks

the push up counter is not working properly what can be the reason

Thanks for your great work! I have some questions about the network structure. By comparing blazepose_full.py with the visualization image of the original tflite model, I found some differences. First, your implementation omitted the "identity_1" output in the original tflite model. Second, the "identity_2" output size is 156, i. e. 4 * (33+6), but the corresponding output size in your implementation is 99, i.e. 3 * 33. Why your implementation is inconsistent with the original model in these aspects? And why the joints output size is 156 in the original model? Many thanks in advance.

hi i am converting LIP CSV file. but here i am getting this error. i preapred same format as you shown dataset.md

File "/home/siva/Documents/wellnesys/tf-blazepose/src/data_loaders/pose_recognition.py", line 90, in load_data
is_pushing_up = int(data["is_pushing_up"])
KeyError: 'is_pushing_up'

你好，你们有blaze-pose的tflite版本的模型吗？

Which model should I use to load the weights for the pretrained weights(.h5) share in the directory in one drive?

When I load the h5 model file, the code reports an error.
File "/home/sa/anaconda3/envs/py38/lib/python3.8/site-packages/tensorflow/python/keras/saving/hdf5_format.py", line 175, in load_model_from_hdf5
raise ValueError('No model found in config file.')
ValueError: No model found in config file.

Traceback (most recent call last):
File "run_video.py", line 93, in
draw = visualize_keypoints(draw, regress_kps[:, :2], visibility=regress_kps_visibility, edges=edges, point_color=(0, 255, 0), text_color=(255, 0, 0))
File "E:\tf-blazepose\src\utils\visualizer.py", line 33, in visualize_keypoints
cv2.line(draw, p1, p2, (0, 0, 255), 2)
cv2.error: OpenCV(4.5.5) 👎 error: (-5:Bad argument) in function 'line'

Overload resolution failed:

• Can't parse 'pt1'. Sequence item with index 0 has a wrong type
• Can't parse 'pt1'. Sequence item with index 0 has a wrong type

I got an error like this. I think it's about CUDA. My device doesn't support CUDA. Help please.

Hi, big fan of your work
I have an issue durnig training heatmap on (lsp/lspet) that the heatmap pck1 is decreasing and the heatmap mae is increasing
What could be the problem?
Thanks!

Hello! I saw your great TF realization. I would be thankful for your help. Now, I'm realizing a similar task for the BlazePose model in TF2. I use another architecture of the model, but it's the same as yours. I'm using this model with pre-trained weights by this graph. And I train this model on our specific dataset, after augmentation we have more 30 000 images.
But, one problem appears during this process: as an output, we have a set of points (coordinates), which is the same for any input, so all coordinates are constant. Perhaps, you have some ideas or direction of thinking for this?
Thank you for your work and help.

Hello.
When I am running the webcam demo with "python run_video.py -c configs/mpii/config_blazepose_mpii_pushup_heatmap_bce_regress_huber.json -m blazepose_pushup_v1.h5 -v webcam --confidence 0.3", an error occurs:

**Traceback (most recent call last):
File "run_video.py", line 45, in
model = trainer.load_model(config, args.model)
File "/home/hkuit155/Desktop/tf-blazepose/src/trainers/blazepose_trainer.py", line 162, in load_model
model.compile()
File "/home/hkuit155/anaconda3/lib/python3.7/site-packages/tensorflow_core/python/training/tracking/base.py", line 457, in _method_wrapper
result = method(self, *args, **kwargs)
File "/home/hkuit155/anaconda3/lib/python3.7/site-packages/tensorflow_core/python/keras/engine/training.py", line 446, in compile
self._compile_weights_loss_and_weighted_metrics()
File "/home/hkuit155/anaconda3/lib/python3.7/site-packages/tensorflow_core/python/training/tracking/base.py", line 457, in _method_wrapper
result = method(self, *args, kwargs)
File "/home/hkuit155/anaconda3/lib/python3.7/site-packages/tensorflow_core/python/keras/engine/training.py", line 1592, in _compile_weights_loss_and_weighted_metrics
self.total_loss = self._prepare_total_loss(masks)
File "/home/hkuit155/anaconda3/lib/python3.7/site-packages/tensorflow_core/python/keras/engine/training.py", line 1691, in _prepare_total_loss
raise ValueError('The model cannot be compiled '
ValueError: The model cannot be compiled because it has no loss to optimize.

It seems that the model needs to be trained before running with webcam. And the error appears when training begins.
Are there any solutions for it? I am looking forward to your reply. Thans you so much.

你好 可不可以提供一下mpii_annotations.json文件或者这个文件在哪里下载，我想试一下mpii数据集，但运行不起来

I tried using mediapipe version to evaluate blazepose on LSP(in which i took some random images),i found that when pose is complex,it doesn't really do a good job compared to other estimation algorithms.
Any thoughts on that?

Can the model infer on CPU or only on GPU?
Can it infer high resolution image or at max, what resolution of the image can be inferred
Can the model perform the 3D pose estimation.
Also let us know what is the Inference FPS.