Things to consider during evaluation

[ybh-yw]

Considerations when evaluating: I want to use t-SNE to visualize your model's results on the DFEW1 fold, but I got a WAR of around 67. Apart from checkpoint = torch.load("../split01_checkpoint-best.pth") and model_without_ddp.load_state_dict(checkpoint['model']), are there any other considerations?

[cyinen]

Please ensure that the testing environment is consistent and use the official DFEW dataset, specifically the clip_224x224_16f version. Then, try again.