Training Yolact with custom dataset - doesn't seem to yield good results

[pankaja0285]

@dbolya -

1. I used your github and followed all the steps -Cloned the repo -downloaded the resnet..50 weights -copied up my train and val images and annotations -modified the config And when I train I see

The model does get generated though, although I don't know the accuracy of it is.. Please let me know what could be the issue and how to improve this. NOTE: Number of images used to train are ~800.

2. Then when I evaluate a whole folder of images (about 140 images), after about 48 images it gives the ndarray error and stops. frame = torch.from_numpy(cv2.imread(path)).cuda().float() TypeError: expected np.ndarray (got NoneType)

Please advice, also if you need any more info from my end let me know. TIA.

[hawk-z]

have you solved this(2) matter

[ningjianfeng]