TIO-IKIM / CellViT

CellViT: Vision Transformers for Precise Cell Segmentation and Classification
https://doi.org/10.1016/j.media.2024.103143
Other
189 stars 27 forks source link

Big svs Prediction #19

Open Transformer-man opened 10 months ago

Transformer-man commented 10 months ago

Hi, when I make a prediction on a svs file of size 191231, 86398 with mpp=0.25, the 128G of RAM just explodes. Any solution for this?

FabianHoerst commented 10 months ago

With this information I can’t help you. Please use the provided bug report template and add all necessary information (function call with terminal command, error messages, part of Programm that fails etc.)

Transformer-man commented 10 months ago

Hello, I didn't make any mistakes. When making predictions, I was able to complete them, but when I finally generated the geojson file, my computer got stuck image

Transformer-man commented 10 months ago

I have set the num worker too small to run, but there are new errors image

Transformer-man commented 10 months ago

Hello. I just used the example you provided, and the model weight used is CellViT-256. I have made many attempts but cannot predict correctly on a machine with 128G memory. Here is the image for the address: https://drive.google.com/file/d/17Wk0LB9dS1pJghlfpqZyh0M2dluVFSSH/view?usp=sharing. Thank you very much for your help.

FabianHoerst commented 10 months ago

We are going to investigate this. It seems that the image is very large with a high amount of cells (>2 Mio). Could be the cause. Is it working for smalle images with less than 1 Mio cells?

464hee commented 10 months ago

However, the sample code already saves the full image clipping locally and loads all the predictions during the fusion process? So it causes a memory overflow?

We are going to investigate this. It seems that the image is very large with a high amount of cells (>2 Mio). Could be the cause. Is it working for smalle images with less than 1 Mio cells?