Closed chartgod closed 5 months ago
Do you mean just loading the ViT pretrained model and making a test without any training epochs?
Yes, that’s right. We would like to test the test dataset by loading the pre-trained model.
Method 1: use torch.load() to load the pretrained model, change the key name (seem to add "model.backbone."?) and use torch.save() to re-save the model.
Method 2: cut down the number of train dataset to only 1 picture, change the learning rate to 0 and train for 1 epoch. Then you can use epoch_1.pth for your need.
Thx
Where is the source code to predict using the pre-trained model?
HRSC2016 STD with Oriented RCNN and ViT-B