MambaVision icon indicating copy to clipboard operation
MambaVision copied to clipboard

Official PyTorch Implementation of MambaVision: A Hybrid Mamba-Transformer Vision Backbone

Results 42 MambaVision issues
Sort by recently updated
recently updated
newest added

Hi, thank you for your excellent work which inspired me a lot. I did some experiments on segmentation task by mmsegmentation toolbox (which version adaptation is really makes me insane...

Great job! Are there any example programs for object detection that are similar to replacing the backbone in YOLO?

Can I use train.py to train on target detection data sets, like LaSOT, GOT10k, etc

![Image](https://github.com/user-attachments/assets/64f71400-0a10-422c-b7a9-34c23b1b2a64) ![Image](https://github.com/user-attachments/assets/a3e54e4e-1d6d-4c67-8dba-6c6a47cdf228) I encountered an issue while installing mambavision==1.1.0, specifically with mamba-ssm==2.2.4. The installation fails during the wheel build process with the following error: NameError: name 'bare_metal_version' is not defined...

Hi, thanks for your amazing work. Could you provide the downstream task code & pretrained weights? Thanks

enhancement

Hi, What is the background of scanning half of channel features instead of all channels? The model splits channels by 2, and half of the channels pass through scanning, and...

hi, very appreciating your excellent work, however, is there any code about converting it to TRT?

Hello, Thank you for sharing your work! As I am training on ImageNet1K, I noticed that memory consumption increases by approximately 254MB with each epoch. If this trend continues, the...

Hi, thanks for your amazing work! I've encountered an issue with NaN losses when using MambaVision to train on CIFAR100 dataset. Could you suggest solutions to solve it? ![Training_loss](https://github.com/user-attachments/assets/7daf50d8-675c-4ee4-9910-5e7c5c0aaa25)

Thank you for your work, it is a truly great innovation. I downloaded your pre-trained weights and achieved outstanding results when fine-tuning them on my own dataset for classification tasks....