Semantic video segmentation is a key challenge for various applications. This paper presents a new model named Noisy-LSTM, which is trainable in an end-to-end manner, with convolutional LSTMs (ConvLSTMs) to leverage the temporal coherency in video frames. We also present a simple yet effective training strategy, which replaces a frame in video sequence with noises. This strategy spoils the temporal coherency in video frames during training and thus makes the temporal links in ConvLSTMs unreliable, which may consequently improve feature extraction from video frames, as well as serve as a regularizer to avoid overfitting, without requiring extra data annotation or computational costs. Experimental results demonstrate that the proposed model can achieve state-of-the-art performances in both the CityScapes and EndoVis2018 datasets.
Download from Cityscapes, set leftimg8bit and gtFine folder under your data_dir. (For Noisy-LSTM training, leftImg8bit_sequence_trainvaltest data is necessary.)
For extra data, download Endovis2018 and set folder under your data_extra.
python train.py --model_name PSPNet --lstm False --noise False --data_dir
python train.py --model_name PSPNet --lstm True --use_pre True --noise False --data_dir
python train.py --model_name PSPNet --lstm True --use_pre True --noise False --noise_type extra --noise_ratio 50 --data_dir --data_extra
This work was supported by Council for Science, Technology and Innovation (CSTI), cross-ministerial Strategic Innovation Promotion Program (SIP), Innovative AI Hospital System (Funding Agency: National Institute of Biomedical Innovation, Health and Nutrition (NIBIOHN)). This work was also supported by JSPS KAKENHI Grant Number 19K10662 and 20K23343.
If you want to use this work, please consider citing the following paper.
@ARTICLE{9382986,
author={B. {Wang} and L. {Li} and Y. {Nakashima} and R. {Kawasaki} and H. {Nagahara} and Y. {Yagi}},
journal={IEEE Access},
title={Noisy-LSTM: Improving Temporal Awareness for Video Semantic Segmentation},
year={2021},
volume={9},
number={},
pages={46810-46820},
doi={10.1109/ACCESS.2021.3067928}}