solawen / fashionai_keypoint_detection_challenge

Code for FashionAI KeyPoint Detection Challenge

FashionAI_KeyPoint_Detection_Challenge

• Code for FashionAI KeyPoint Detection
• Score of this code : 5.59%
• Team rank 58/2322 at 1st round competition and 32/2322 at 2nd round competition
• Mainly base on Cascaded Pyramid Network for Multi-Person Pose Estimation.

Folder Structure

• train: store training images and annotations
• model: store trained models
• test : store testing images
• src: store some of the source code.
  src/CPN_ONE_data_gen.py: data generator.
  src/CPN_ONE_model.py: model definition.
• predict.py : predict keypoints from images.
• preprocessing.py : generate .mat file for training model.
• train.py : training model.

Usage

1.Run the following command to generate .mat file

python preprocessing.py

2.Run the following command to train

python train.py --gpus=1

3.Modify the model path in predict (model_to_load), then run the following command to get predict .csv file.

python predict.py

Requirements

• python (3.5.2)
• keras (2.1.5)
• numpy (1.14.2)
• opencv-python (3.4.0.12)
• pandas (0.22.0)
• scipy (1.0.1)
• tensorflow-gpu (1.2.0)

fashionai_keypoint_detection_challenge's Issues

Missing file CPN_model

Trying to run the training script, return the following error.
Looks like a missing src.CPN_model file.

from src.CPN_model import CPN2
ImportError: No module named 'src.CPN_model'

share your trained model?

Hello @solawen

Thank for sharing your work!

Could you share the latest trained model?

I have stumbled upon your repository while looking for clothes detectors.
It would be helpful for me to run predictions with your model for figuring out which direction I should work on.

Thanks in advance.

train into one moel or train one by one?

thanks for sharing , I want to know if this code could put different categories of clothes into one model? If it works ,how can I make it? I am looking forward to your reply ,thank you very much

Can you share your data?

Hey!
Can you provide a google drive or a similar link to the data that you used (because the ones on the site of the challenge are not downloadable anymore)? Or can you share your trained model?