Qsingle / Megvision

Some models implemented by megengine

Megvision

The MegEngine vision of the models we used in our experiments, hope this repository can help with you.

Usage

Sample to import the DPT based on the SAM.

import megengine as mge
from megvision.model.segmentation.dpt import DPT
from megengine import random

model = DPT(real_img_size=1024, arch="sam_vit_b", img_size=1024, checkpoint=None)
model.eval()
x = random.normal(size=(1, 3, 1024, 1024))
out = model(x)
print(out.shape)

You can also see the file iostar_train.py as a sample for training.

Supported Model List

• Classification

  • AlexNet: One weird trick for parallelizing convolutional neural networks
  • VGGNet: Very Deep Convolutional Networks For Large-Scale Image Recognition
  • ResNets: Deep Residual Learning for Image Recognition
    • WideResNets: Wide Residual Networks
    • ResNeXt: Aggregated Residual Transformation for Deep Neural Networks
    • SENets: Squeeze-and-Excitation Networks
    • ResNeSt: ResNeSt: Split-Attention Networks

• Segmentation:

  • SuperVessel SuperVessel: Segmenting High-resolution Vessel from Low-resolution Retinal Image
  • PSPNet Pyramid Scene Parsing Network
  • SCSNet:SCS-Net: A Scale and Context Sensitive Network for Retinal Vessel Segmentation
  • ESPNets:
    • V2:ESPNetv2: A Light-weight, Power Efficient, and General Purpose Convolutional Neural Network
  • DeepLab:
    • V3:Rethinking Atrous Convolution for Semantic Image Segmentation
    • V3Plus:Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation
  • CE-Net:CE-Net: Context Encoder Network for 2D Medical Image Segmentation
  • CS-Net: CS2-Net: Deep learning segmentation of curvilinear structures in medical imaging
  • PFSeg: Patch-free 3D Medical Image Segmentation Driven by Super-Resolution Technique and Self-Supervised Guidance
  • DPT: Vision Transformers for Dense Prediction
  • SAM: Segment Anything
    • Thanks for the MegEngine group to provide the MegEngine-SAM implementation.