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

By Vladimir I. Iglovikov, Alexey A. Shvets. Computer-Aided Analysis of Gastrointestinal Videos, 2018

677 citations recorded by the source corpus.

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 this https URL. 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.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/0269073e07faaafa4df7b3d7ddac96e6dccf853d

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

Repurposing GANs for One-Shot Semantic Part Segmentation (2021, background): “Many approaches learn these representations from solving one task and employ them to help improve the performance on another task [12], [13], [14], [15], [16], [17].”
Deepfacelab: Integrated, flexible and extensible face-swapping framework (2020, methodology): “We employ a ﬁne-grained Face Segmentation network (TernausNet [8]) on top of ( aligned src or aligned dst ), through which a face with either hair, ﬁngers, or glasses could be segmented exactly.”
JCS: An Explainable COVID-19 Diagnosis System by Joint Classification and Segmentation (2020, background): “It is further extended to 3D U-Net [59], TernausNet [60], and U-Net ++ [61] with promising performance on versatile image segmentation tasks.”
U-NetPlus: A Modified Encoder-Decoder U-Net Architecture for Semantic and Instance Segmentation of Surgical Instruments from Laparoscopic Images (2019, result): “From the above-mentioned papers, only one seems to have achieved results comparable to ours [9], but it still suffers from several artifacts, which we have been able to further mitigate some of these artifacts using our proposed method.”
2017 Robotic Instrument Segmentation Challenge (2019, positive): “Further improvements take into account encoders such as VGG, ResNets etc. pre-trained on ImageNet [28].”
CHAOS Challenge - Combined (CT-MR) Healthy Abdominal Organ Segmentation (2020, background): “TernausNet (U-Net with VGG11 backbone (Iglovikov and Shvets, 2018)), LinkNet34 (Shvets et al.”
Semi-Supervised Segmentation of Salt Bodies in Seismic Images using an Ensemble of Convolutional Neural Networks (2019, methodology): “Hence, we opt to use an Imagenet pretrained CNN as the backbone for the encoder [18].”
Breast Tumor Cellularity Assessment using Deep Neural Networks (2019, methodology): “[28, 29] used batch normalization [30] and exponential linear unit (ELU) as the primary activation function and an ImageNet pre-trained VGG-11 network [31] as an encoder.”
A Multi-Organ Nucleus Segmentation Challenge (2020, methodology): “Among the remaining, four teams used their own custom models and architectures, and one each used VGGNet [40], Deep Layer Aggregation [45], PANet [46], and TernausNet [47].”