search

mks0601 / PoseFix_RELEASE

Official TensorFlow implementation of "PoseFix: Model-agnostic General Human Pose Refinement Network", CVPR 2019
MIT License
327 stars 64 forks source link

Find data #44

Open wangdong0556 opened 3 years ago

wangdong0556 commented 3 years ago

Thank you for providing this function. Where can I find name_of_input_pose. Json and result.json ? Can you provide the download address

mks0601 commented 3 years ago

you can use any 2D pose estimation methods to obtain name_of_inptu_pose.json. For example, here

wangdong0556 commented 3 years ago

Thank you for your reply, but I can't jump when I click 'here'. In addition, I saw in 'TF-Simplehumanpose' that someone asked how to find 'human_detection. Json', but it showed 'Closed this on 6 Feb 2019', 'name_of_input_pose' and 'name_of_inptu_pose' are the same?Is' result.json '2D result of' name_of_input_pose. Json '?

metehkaya commented 2 years ago

Can the file you provided here be directly used during training or testing? I guess, it is just a part of MS COCO json file, right? Because according to video, it seems that MS COCO compatible json file is actually larger.

mks0601 commented 2 years ago

That file is 2D pose estimation result of other 2D pose estimator, not a part of MSCOCO dataset. You can train and test on that file, but the trained model will not be model-agnostic refiner, while the PoseFix is.

metehkaya commented 2 years ago

I am aware that the data you provided is not a part of MSCOCO dataset. Actually what I ask is that should that file follow MS COCO format completely. I guess not, because you don't make COCO(name_of_inptu_pose.json) at input_pose_load function like you did at COCO(self.test_annot_path) at load_annot function.

mks0601 commented 2 years ago

Sure. The file format of that file follows 2D pose estimation output format, used to evaluate it by COCO evaluation API. The file format is completely different from COCO(self.test_annot_path)