liuruiyang98/jittor-mlp

Unofficial Implementation of MLP-Mixer, gMLP, resMLP, Vision Permutator, S2MLP, S2MLPv2, RaftMLP, HireMLP, ConvMLP, AS-MLP, SparseMLP, ConvMixer, SwinMLP, RepMLPNet, WaveMLP, MorphMLP, DynaMixer, MS-MLP, Sequencer2D in Jittor and PyTorch! Now, Rearrange and Reduce in einops.layers.jittor are support!! trunc_normal_ is supported for Jittor!! MLP Paper is uploaded!

Hi , I’m Ruiyang Liu 👋

Things about me:

• 👶 Master's Student in computer vision.
• 🔭 Insterested in both traditional computer vision and deep learning.
  • Security fields including edge detection, basic image feature, what a neutal network learning, etc...
• ⚡ Currently focusing on discrete signal gradient calculation.
• 🍎 Blog: https://liuruiyang98.github.io/
• 🍎 CSDN: https://blog.csdn.net/baidu_36913330

⚡ GitHub Stats:

jittor-mlp's Issues

Might miss a padding?

Hi, good to find your implementation!

Here, I wonder whether miss a padding=patch_size//2

Jittor-MLP/models_pytorch/conv_mixer.py

Line 18 in 68d5877

nn.Conv2d(3, dim, kernel_size=patch_size, stride=patch_size),

Best

Comparison with reported results

Thank you for coding this.
Have you compared against the reported results in the papers?

code problem

Hi, thank you for your great work! I think the value of dim should be 2.

Jittor-MLP/models_pytorch/hire_mlp.py

Line 113 in b86656b

self.cross_region_restoreH = CrossRegion(step = -cross_region_step, dim = 3)

question

Thank you for coding this！
I want to know why this code is written like this and what the padding operation acts on.

Jittor-MLP/models_pytorch/morph_mlp.py

Line 23 in b86656b

class MorphFC(nn.Module):

R2 -MLP: Round-Roll MLP Architecture for Multi-View 3D Object Recognition

How difficult would it be to implement this one?

https://arxiv.org/abs/2211.11085

Should be very similar to S2-MLP , right ?