search

bochenxie / EISNet

[IEEE TMM 2024] EISNet: A Multi-Modal Fusion Network for Semantic Segmentation with Events and Images
MIT License
7 stars 0 forks source link

Code release requirement #1

Closed lyu-yx closed 1 month ago

lyu-yx commented 3 months ago

Dear Bochen,

Thanks for your great work! I am wondering when will the code be released?

Best, Yixuan

Candy-Crusher commented 1 month ago

Any updates?

bochenxie commented 1 month ago

Dear Bochen,

Thanks for your great work! I am wondering when will the code be released?

Best, Yixuan

Dear Yixuan @lyu-yx,

Thank you for your interest in our work! The source code of this project has been released. For project guidelines and model weights, I'll upload them in the next few days.

Besides, Happy Mid-Autumn Festival!

Candy-Crusher commented 1 month ago

image "Since the original papers of EDCNet-S2D, CMX, and CMNeXt do not evaluate these two datasets, we re-implement them based on the source code with a 3-channel voxel grid [15] as event representation." May I ask if you also use DSEC-Semantic train from scratch in reproducing the three networks EDCNet-S2D, CMX, and CMNeXt?

bochenxie commented 1 month ago

"Since the original papers of EDCNet-S2D, CMX, and CMNeXt do not evaluate these two datasets, we re-implement them based on the source code with a 3-channel voxel grid [15] as event representation." May I ask if you also use DSEC-Semantic train from scratch in reproducing the three networks EDCNet-S2D, CMX, and CMNeXt?

@Candy-Crusher Thank you for your inquiry. Not from scratch. At the training stage, the encoders of these three models are loaded with ImageNet pre-trained weights. As for the decoders, they are randomly initialized.