[Docs] Testing pre-train model on other dataset

[Mr-Cronk]

📚 The doc issue

I'm just starting my work with mmpose and getting acquainted with all the tools of this solution. I would like to test pre-trained models on different datasets (for example, perform a test of the pre-trained scnet101_coco_384x288 model on the mpii dataset) in order to conduct analysis and some calculations in the future. I changed the paths to the data and annotations in the configuration file, however, I did not get the expected result and only received an error about data inconsistency. I apologize in advance, but could you please help me find a solution to my problem?

Suggest a potential alternative/fix

A good idea would be to add a paragraph in the "Training and testing" section with information on how to do this.

[Tau-J]

Hi @Mr-Cronk , please follow our issue template to provide necessary information so that we can give you more help.

[Mr-Cronk]

Hi @Mr-Cronk , please follow our issue template to provide necessary information so that we can give you more help.

I apologize if my comment misled you. I want to run testing using the pretrained model https://github.com/open-mmlab/mmpose/blob/5a3be9451bdfdad2053a90dc1199e3ff1ea1a409/configs/body_2d_keypoint/topdown_heatmap/coco/td-hm_scnet101_8xb48-210e_coco-384x288.py with https://download.openmmlab.com/mmpose/top_down/scnet/scnet101_coco_384x288-0b6e631b_20200709.pth on the MPII dataset, as per the documentation, using tools/test.py. I would like to perform similar tests with other models and datasets. How can I do this without changing the annotations of the MPII dataset to Coco?

[Ben-Louis]

You can refer to custom testing to perform keypoint order conversion during testing