Daily paper reading record

SSAP: Single-Shot Instance Segmentation With Affinity Pyramid

date: 2019/10/29

categories: instance segmentation

summary: This work has proposed a single-shot proposal-free instance segmentation method, which requires only one single pass to generate instances. Our method is based on a novel affinity pyramid to distinguish instances, which can be jointly learned with the pixel-level semantic class labels using a single backbone network.

pipeline:

Real-time self-adaptive deep stereo

date: 2019/10/30

categories: deep stereo

summary: The proposed online unsupervised fine-tuning approach can successfully tackle the domain adaptation issue for deep end-to-end disparity regression networks. For applications in which inference time is critical, we have proposed MADNet, a novel network architecture, and MAD, a strategy to effectively adapt it online very efficiently. We believe this to be key to practical deployment of these potentially groundbreaking deep learning systems in many relevant scenarios.

pipeline:

CCNet: Criss-Cross Attention for Semantic Segmentation

date: 2019/10/30

categories: semantic segmentation

summary: In this paper, we have presented a Criss-Cross Network (CCNet) for semantic segmentation, which adaptively captures long-range contextual information on the criss-cross path. To obtain dense contextual information, we introduce recurrent criss-cross attention module which aggregates contextual information from all pixels

pipeline:



core idea:

DeepMVS-Learning Multi-view Stereopsis

date: 2019/11/5

categories: Multi-view reconstruction

summary: With DeepMVS, we demonstrate the feasibility of learning Mulit-View Stereopsis with a convolutinoal neural network, and show that learning-based approaches can overcome the weaknesses of conventional algorithms.

pipeline:



link: DeepMVS-Learning Multi-view Stereopsis

DPSNET: END-TO-END DEEP PLANE SWEEP STEREO

date: 2019/11/6

categories: Multi-view reconstruction

summary: We developed a multiview stereo network whose design is inspired by best practices of traditional non-learning-based techniques. Moreover, we propose a context-aware cost aggregation method that leads to improved depth regression without any post-processing.

pipeline:



link: DPSNET: END-TO-END DEEP PLANE SWEEP STEREO

Asymmetric Non-local Neural Networks for Semantic Segmentation

date: 2019/11/10

categories: Semantic segmentation

summary: In this paper, we propose an asymmetric non-local neural network for semantic segmentation. The core contribution of asymmetric non-local neural network is the asymmetric pyramid non-local block, which can dramatically improve the efficiency and decrease the memory consumption of non-local neural blocks without sacrificing the performance. Besides, we also propose asymmetric fusion non-local block to fuse features of different levels. The asymmetric fusion non-local block can explore the long range spatial relevance among features of different levels

pipeline: