Skip to content

Tensorflow implementation of UNet on surgical Instrument dataset from laparoscopic videos

Notifications You must be signed in to change notification settings

DebeshJha/UNet-insturment-segmentation

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

12 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

U-Net Architecture for Surgical Image Segmentation (ROBUST-MIS-Challenge-dataset)

U-Net is a renowned convolutional neural network (CNN) designed primarily for biomedical image segmentation.

Kindly note that the figure is taken from the original paper.

🔍 Key Features

  • Symmetrical Design: Comprising an encoder (contracting path) and a decoder (expansive path).
  • Skip Connections: Retain spatial context by bridging encoder and decoder.
  • Efficient and Precise: Requires fewer training samples but offers high precision.

Structure

  1. Encoder:

    • Two 3x3 convolutions followed by a ReLU and a 2x2 max pooling.
    • Feature channels double after each pooling.
  2. Bottleneck:

    • Two 3x3 convolutions with ReLU activation.
  3. Decoder:

    • Upsampling and a 2x2 transposed convolution.
    • Concatenation with skip-connected feature map from the encoder.
    • Two 3x3 convolutions with ReLU.
  4. Output:

    • 1x1 convolution to map to the number of classes.

Applications

  • Original: Biomedical image segmentation.
  • Extended: Any Semantic segmentation tasks.

For a deeper dive, check out the link_to_paper).

We have implemented U-Net on the Binary instrument segmentation task from the ROBUST-MIS Challenge dataset.

Dataset

We have used a binary segmentation dataset from the ROBUST-MIS Challenge. You can down the dataset from here.

Results

Discussion

We got the best results with DDANet on this dataset.

The official link to the DDANet code can be found here.

References

Please cite our work Exploring Deep Learning Methods for Real-Time Surgical Instrument Segmentation in Laparoscopy if you find it useful.