Skip to content

Latest commit

 

History

History
22 lines (14 loc) · 1.55 KB

TROUBLESHOOTING.md

File metadata and controls

22 lines (14 loc) · 1.55 KB

Troubleshooting

Below is a list of common errors and their possible solutions:

CUDA related errors: 

d2.utils.memory Attempting to copy inputs of <function paste_masks_in_image at ...> to CPU due to CUDA OOM

If this error occurs during the sampling of the most uncertain images, it might indicate that your GPU has memory issues when sampling the images. A possible solution is to reduce the number of Monte-Carlo iterations (mcd_iterations) and/or the pool size (pool_size) in your configuration file (maskAL.yaml).

Error in function check_json_presence: 

active_learning.sampling.prepare_dataset - ERROR - Error in function check_json_presence: 
FileNotFoundError(2, 'No such file or directory')

This error can occur when the network_config or pretrained_weights is not specified correctly in the maskAL.yaml file (when auto_annotate is set to True). The Detectron2-code assumes that you are using relative file-paths instead of absolute file-paths. For example: the pretrained COCO-weights always need to be specified as COCO-InstanceSegmentation/[...].yaml and not ~/maskal/configs/COCO-InstanceSegmentation/[...].yaml.

If you are using the Windows operating system then specify the yaml file with a forward slash (COCO-InstanceSegmentation/[...].yaml), otherwise an error will be thrown. When you want to use a specific pretrained weights-file under Windows then use two backward slashes (for example: C:\\Users\\PieterBlok\\Documents\\maskal\\weights\\my_own_weights.pth).