search

dbolya / yolact

A simple, fully convolutional model for real-time instance segmentation.
MIT License
5k stars 1.33k forks source link

Can't train on custom dataset: DataLoader worker process 0. AND object of type 'int' has no len() #626

Open Atonbom opened 3 years ago

Atonbom commented 3 years ago

Hi,

I'm trying to get YOLACT(++) to work with a custom dataset using Google Colab because I don't have acces to a powerfull GPU myself. When I run train.py it loads the .json files from the config.py and says "Begin training". But then I get the following error and I have no clue where to start looking or what I'm doing wrong. I googled for both errors but I can't find any specific information on how to solve the issue.

image

Could someone please help me out here? If not I will go back to Mask-RCNN but I would like to get this working it annoys me

kidpaul94 commented 3 years ago

I presume that you are trying to do transfer learning with your own config and dataset. Try to search how to do transfer learning for Yolact++. https://github.com/dbolya/yolact/issues/36 will tell you how to do fine tuning. If you want to only modify the last soft-max layer, https://github.com/dbolya/yolact/issues/334 may be helpful.

On top of this, you need to call pretrained weights when doing the transfer learning (e.g., yolact_resnet50_54_800000.pth)