0

I am trying to train a custom data for image segmentation with Detectron2, but I have an issue while using the config files (like mask_rcnn_R_50_FPN_3x.yaml).

Here is the the configuration that I use for training:

cfg = get_cfg()
cfg.MODEL.DEVICE = "cpu"
cfg.DATASETS.TRAIN = ("category_train",)
cfg.DATASETS.TEST = ()

cfg.merge_from_file(model_zoo.get_config_file("COCO-InstanceSegmentation/mask_rcnn_R_50_FPN_3x.yaml"))
cfg.MODEL.WEIGHTS = "COCO-InstanceSegmentation/mask_rcnn_R_50_FPN_3x.yaml"

cfg.DATALOADER.NUM_WORKERS = 0
cfg.SOLVER.IMS_PER_BATCH = 8
cfg.SOLVER.BASE_LR = 0.00025  # pick a good LR

cfg.SOLVER.MAX_ITER = 25000
cfg.SOLVER.STEPS = []
cfg.MODEL.ROI_HEADS.BATCH_SIZE_PER_IMAGE = 512
cfg.MODEL.ROI_HEADS.NUM_CLASSES = 6

The part where I get error is cfg.merge_from_file() part. If I remove it and cfg.MODEL.WEIGHTS, then the code works and model successfully trains the object detection model. I need image segmentation, so I need to use these configurations.

However, using them gives the following error:

FileNotFoundError: [Errno 2] No such file or directory: 'datasets\\coco/annotations/instances_train2017.json'

I looked in the Detectron2 issues and someone also had the same error, and the solution was to put the following in code:

cfg.DATASETS.TEST = ()

However, I had already done that, and it did not help me. Does anyone have any ideas on why this could not work?

Kamran Gasimov
  • 36
  • 3
  • Please clarify your specific problem or provide additional details to highlight exactly what you need. As it's currently written, it's hard to tell exactly what you're asking. – Community Oct 08 '22 at 09:17

1 Answers1

0

check the path to the file instances_train2017.json, there is a problem with the path to that file, datasets\\coco/annotations/instances_train2017.json is not the correct path. please check the configuration within this file COCO-InstanceSegmentation/mask_rcnn_R_50_FPN_3x.yaml I hope you will see that line inside.

Ismaili Mohamedi
  • 906
  • 7
  • 15