search

jackyjsy / CVPR21Chal-SLR

This repo contains the official code of our work SAM-SLR which won the CVPR 2021 Challenge on Large Scale Signer Independent Isolated Sign Language Recognition.
Creative Commons Zero v1.0 Universal
205 stars 50 forks source link

How to get correct 'train_val_labels.pkl' ? #23

Closed LiuyhLinda closed 2 years ago

LiuyhLinda commented 2 years ago

Thanks for your project. I'm very interested in the network you designed. But when goes to 'test_joint.yaml', I can not get the 'train_val_labels.pkl'. The 'train_val_labels.pkl' I merged from 'train_label.pkl' and 'val_gt.pkl' is wrong. The code is following:

import pickle with open('/media/lyh/data/AUTSL/Autsl/skeleton_data/train_label.pkl', 'rb') as fr_1: data_1 = pickle.load(fr_1) with open('/media/lyh/data/AUTSL/Autsl/skeleton_data/val_gt.pkl', 'rb') as fr_2: data_2 = pickle.load(fr_2) data = [data_1, data_2] with open('/media/lyh/data/AUTSL/Autsl/skeleton_data/train_val_labels.pkl', 'wb') as fr_3: pickle.dump(data, fr_3)
​### Would you mind help me to get correct 'train_val_labels.pkl'? I am looking forward to your reply. Thanks in advance.

jackyjsy commented 2 years ago

Hi there, the train_val_labels.pkl is only used when attending the challenges where we were allowed to finetune our model on validation set. When you test the model using test_joint.yaml, it only specifies the training loader, so not used in testing. If you only want to test on the test set, you can change it to train_labels.pkl, which won't affects your model accuracy on the test set. I have made a new commit address this issue.

LiuyhLinda commented 2 years ago

It is working! Thanks a lot!