search

ziqi-jin / finetune-anything

Fine-tune SAM (Segment Anything Model) for computer vision tasks such as semantic segmentation, matting, detection ... in specific scenarios
MIT License
787 stars 56 forks source link

about TorchVOC finetuning results #40

Open lywang76 opened 1 year ago

lywang76 commented 1 year ago

Hello, I ran your codes on TorchVOC dataset.

After 200 iterations, i got the following results. image

It seems mIOU is very low. Did you get similar results? I used your default configuration file.

Thanks, Liya

TimandXiyu commented 1 year ago

I also noticed the convergence is slow enough that there might be some issue with the training config. I found the mIoU around 20 after 7000 iterations, which is kind of shocking for model like SAM after some intense finetuning.

jiaweichaojwc commented 11 months ago

@TimandXiyu Hello! Through my tests, I've found that the loss is challenging to decrease, mostly oscillating, and the mIoU is quite low. I feel that the mask adapter approach may not be suitable for fine-tuning. Have you recently tested the code repository, and did you get similar results to mine?