search

cure-lab / DeciWatch

[ECCV 2022] Official implementation of the paper "DeciWatch: A Simple Baseline for 10x Efficient 2D and 3D Pose Estimation"
Apache License 2.0
177 stars 15 forks source link

Is action_name required for the dataset? #21

Open cacticaxas opened 1 year ago

cacticaxas commented 1 year ago

Hello,

I'm trying to implement a dataset that does not have any action_name. In the documentation provided (https://github.com/cure-lab/DeciWatch/blob/main/doc/data.md) it states that the data should be given in the following format of [action_name]/[sequence_name]/[image_id]. Can the action_name be omitted or must it be provided for my custom dataset?

juxuan27 commented 1 year ago

You can simply omit it but this may cause some change of the code in visualization part (if you use the visualization of Human3.6M). But if you simply want to train and evaluate, this will not cause any problem.

cacticaxas commented 1 year ago

I see. Thank you for the explanation. I have, however, a follow-up question.

In my custom dataset, I have more than one pose annotations for each person in the frame. However, I noticed in the .npz files format that was provided, only one annotation is assigned to each frame eventhough there is more than one person in the frame. In this case I am referring to the "joints_2d" field in Sub-JHMDB dataset, where the shape of each sequence is the corresponding_sequence_length(152).

Is it possible to include more than one annotations per frame for multi-person human pose estimation?

juxuan27 commented 1 year ago

Sorry for the late reply. You may change the shape to corresponding_sequence_length(302), where the 30 is the concat of 2 15 (in case you have 2 person). However, you may need some effort to change the visualization code if you do so.