TernausNet: U-Net with VGG11 Encoder Pre-Trained on ImageNet for Image Segmentation

Abstract

Pixel-wise image segmentation is demanding task in computer vision. Classical U-Net architectures composed of encoders and decoders are very popular for segmentation of medical images, satellite images etc. Typically, neural network initialized with weights from a network pre-trained on a large data set like ImageNet shows better performance than those trained from scratch on a small dataset. In some practical applications, particularly in medicine and traffic safety, the accuracy of the models is of utmost importance. In this paper, we demonstrate how the U-Net type architecture can be improved by the use of the pre-trained encoder. Our code and corresponding pre-trained weights are publicly available at https://github.com/ternaus/TernausNet. We compare three weight initialization schemes: LeCun uniform, the encoder with weights from VGG11 and full network trained on the Carvana dataset. This network architecture was a part of the winning solution (1st out of 735) in the Kaggle: Carvana Image Masking Challenge.

5 pages, 4 figures

References in corpus (3)

• Very Deep Convolutional Networks for Large-Scale Image Recognition
• Satellite Imagery Feature Detection using Deep Convolutional Neural Network: A Kaggle Competition
• U-Net: Convolutional Networks for Biomedical Image Segmentation

Cited by in corpus (30)

• Global Guidance Network for Breast Lesion Segmentation in Ultrasound Images
• 2017 Robotic Instrument Segmentation Challenge
• Automated Cardiothoracic Ratio Calculation and Cardiomegaly Detection using Deep Learning Approach
• RAUNet: Residual Attention U-Net for Semantic Segmentation of Cataract Surgical Instruments
• DeepFlash: Turning a Flash Selfie into a Studio Portrait
• Theoretical analysis and experimental validation of volume bias of soft Dice optimized segmentation maps in the context of inherent uncertainty
• Joint Iris Segmentation and Localization Using Deep Multi-task Learning Framework
• U-NetPlus: A Modified Encoder-Decoder U-Net Architecture for Semantic and Instance Segmentation of Surgical Instrument
• SpaceNet 6: Multi-Sensor All Weather Mapping Dataset
• Surgical Visual Domain Adaptation: Results from the MICCAI 2020 SurgVisDom Challenge
• Unsupervised Medical Image Segmentation with Adversarial Networks: From Edge Diagrams to Segmentation Maps
• Deep Landscape Features for Improving Vector-borne Disease Prediction
• Deep Learning-based Aerial Image Segmentation with Open Data for Disaster Impact Assessment
• Effect of the output activation function on the probabilities and errors in medical image segmentation
• TricycleGAN: Unsupervised Image Synthesis and Segmentation Based on Shape Priors
• Applying Knowledge Transfer for Water Body Segmentation in Peru
• BLESER: Bug Localization Based on Enhanced Semantic Retrieval
• Farmland Parcel Delineation Using Spatio-temporal Convolutional Networks
• Switching Loss for Generalized Nucleus Detection in Histopathology
• MSDU-net: A Multi-Scale Dilated U-net for Blur Detection
• Data-Free Adversarial Perturbations for Practical Black-Box Attack
• Human Driver Behavior Prediction based on UrbanFlow
• Transfer Learning in Visual and Relational Reasoning
• A Study of Domain Generalization on Ultrasound-based Multi-Class Segmentation of Arteries, Veins, Ligaments, and Nerves Using Transfer Learning
• Solving Traffic4Cast Competition with U-Net and Temporal Domain Adaptation
• Towards Comparative Physical Interpretation of Spatial Variability Aware Neural Networks: A Summary of Results
• Improving Building Segmentation for Off-Nadir Satellite Imagery
• Network-Agnostic Knowledge Transfer for Medical Image Segmentation
• Automatic Polyp Segmentation using U-Net-ResNet50
• IntrinSeqNet: Learning to Estimate the Reflectance from Varying Illumination