I have been following a tutorial on training a segmentation model on a custom dataset, but it refuses to make any progrees in training the model.

I am trying to fine-tune mit-b0 segmentation model on satellite images to segment rice paddies on RTX 2070 with 8 GB VRAM but I get CUDA out of memory at the beginning of the first epoch. I believe I have some memory allocation problem, please let me know what is the issue. I think I should be able to fit the b0 on my GPU

I’m trying to build a UNet Segmentor for an aerial image of vegetation w/ pytorch. The UNet then gets exported to an ONNX/GIS_ONNX to be usable in the Deepness-Plugin of QGIS. The Unet-Basecode and the training script are mostly derived/copied from Mostafa Wael on toward data science. The Dataloader is a Custom-Build upon the torch.utils.dataloader class. Training data is fed in tile-pairs of cut ortho mosaic and corresponding class mask (512×512).

Helllo I’m currently trying to install MMsegmentation from this official tutorial github:
https://github.com/open-mmlab/mmsegmentation