human-like figure detection and pose estimation model

Question

human-like figure detection and pose estimation model

LongJi12 opened this issue 8 months ago · comments

I would like to use the dataset you provided to train the human body detection model and pose estimation model used in the code. Can you provide code for these two models？

hjessmith · Answer 1 · Fri Oct 13 2023 00:21:39 GMT+0800 (China Standard Time)

Hello @LongJi12. The models were trained using the OpenMMLab framework. If you download the .mar files and unzip them, you will find the config files OpenMMLab uses to specify model architecture.

LongJi12 · Answer 2 · Fri Oct 13 2023 23:47:22 GMT+0800 (China Standard Time)

I have successfully extracted the. mar file and reviewed the relevant model content, but I still have difficulties during training. The downloaded dataset does not include the 2500 manually selected image materials mentioned in the paper and similar training materials_ Size-180000_ Dev.json, train_ Size-180000_ Can you provide ? At 2023-10-13 00:21:50, "hjessmith" ***@***.***> wrote: Hello @LongJi12. The models were trained using the OpenMMLab framework. If you download the .mar files and unzip them, you will find the config files OpenMMLab uses to specify model architecture. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: ***@***.***>

hjessmith · Answer 3 · Fri Oct 13 2023 23:55:50 GMT+0800 (China Standard Time)

Let me look around for them. I no longer have access to the environment I used to train, so I'll have to look around for it. Are you attempting to compare against it as a baseline or benchmark?

LongJi12 · Answer 4 · Sat Oct 14 2023 00:02:44 GMT+0800 (China Standard Time)

Let me look around for them. I no longer have access to the environment I used to train, so I'll have to look around for it. Are you attempting to compare against it as a baseline or benchmark?

I want to use it as a benchmark for comparison

javier-pkg-mda · Answer 5 · Tue Nov 21 2023 21:18:42 GMT+0800 (China Standard Time)

@hjessmith Did you manage to get these datasets used by the demo? I'm getting way worse outcomes using just dataset given in this repo.

hjessmith · Answer 6 · Wed Mar 27 2024 02:50:37 GMT+0800 (China Standard Time)

After several unsuccessful searches, I've concluded that the list of 2500 manually selected images used to generate the smaller dataset didn't survive the environment transfer. I was able to find an incomplete list of the manually selected images, however, which I'll link here. Hopefully that can be of some use: https://drive.google.com/file/d/17fXv9NithNiFmmn_UmbMvWvJNZYL8zXv/view?usp=sharing