Error running MaskRCNN inference after custom training - batch size

I am trying to train MaskRCNN using TLT. The training with the COCO data set was successful. Training with a custom data set was fine and the training ended with ‘success’ status. The evaluation was also fine. At the inference stage an error message appears saying “The batch size should not be larger than the number of samples”, and then the process run failed.

I will appreciate your help

Please check if you have png files. PNG file is not supported.


  • Input size : C * W * H (where C = 3, W > =128, H >=128 and W, H are multiples of 32)
  • Image format : JPG
  • Label format : COCO detection
1 Like

It worked. Thanks for your support. The training was with PNG files.

TLT 3.0-py3 version can support png files for inference.

1 Like