Maskrcnn-benchmark: How do trained models are loaded for inference?

[maxsenh]

❓ Questions and Help

Hi

I trained my own model while using the pre-trained Imagenet model. Now, I have trouble understanding how my model is loaded for inference? Is first the model found in OUTPUT_DIR loaded and then the one from MODEL.WEIGHT ?

        save_dir = cfg.OUTPUT_DIR
        checkpointer = DetectronCheckpointer(cfg, self.model, save_dir=save_dir)        
        _ = checkpointer.load(cfg.MODEL.WEIGHT)

Thanks in advance

[fmassa]

If there is a last_checkpoint file in OUTPUT_DIR, it will take the file present there.
If there isn't any file named last_checkpoint in OUTPUT_DIR, it will load the weights from MODEL.WEIGHT.

[JoyHuYY1412]

If there is a last_checkpoint file in OUTPUT_DIR, it will take the file present there.
If there isn't any file named last_checkpoint in OUTPUT_DIR, it will load the weights from MODEL.WEIGHT.

if I just write,

save_dir= cfg.OUTPUT_DIR
checkpointer = DetectronCheckpointer(cfg, self.model, save_dir=save_dir) 

If there isn't any file named last_checkpoint in OUTPUT_DIR, it will start from the scratch, right?