Reading List(Object Detection Related)

Here are my notes and thoughts for papers.

Anchor Free

1. CornerNet: Detecting Objects as Paired Keypoints(CornerNet) 👀
2. Bottom-up Object Detection by Grouping Extreme and Center Points(ExtremeNet)👀
3. Feature Selective Anchor-Free Module for Single-Shot Object Detection(FSAF)
4. FCOS: Fully Convolutional One-Stage Object Detection(FCOS)
5. Single-Shot Refinement Neural Network for Object Detection(RefineDet)👀
6. Revisiting Feature Alignment for One-stage Object Detection(AlignDet)
7. RepPoints: Point Set Representation for Object Detection(RepPoints、RPDet)
8. Bridging the Gap Between Anchor-based and Anchor-free Detection via Adaptive Training Sample Selection(ATSS)

Rescoring for better mAP

1. Acquisition of Localization Confidence for Accurate Object Detection(IoUNet)
2. Mask Scoring RCNN(MS-RCNN)
3. Learning to Rank Proposals for Object Detection(NMS-LTR)

Better Alignment

1. Cascade R-CNN: Delving into High Quality Object Detection(Cascade R-CNN)
2. Revisiting Feature Alignment for One-stage Object Detection(AlignDet)
3. Cascade RPN: Delving into High-Quality Region Proposal Network with Adaptive Convolution(Cascade RPN)
4. Shape-aware Feature Extraction for Instance Segmentation(Mask Refining R-CNN)
5. Scale-Aware Trident Networks for Object Detection(TridentNet, better anchor align for size?)
6. Region Proposal by Guided Anchoring(GA-RPN, deformable anchor)

Knowledge Distillation for Object Detection

1. Distilling the knowledge in a Neural Network(start of KD)
2. Mimicking Very Efficient Network for Object Detection(ROI Teaching)
3. Learning Efficient Object Detection Models with Knowledge Distillatio(Hint + Reg + Cls distillation)
4. Distilling Object Detection with Fine-grained Feature Imitation(Hint with Anchor Mask)
5. Consistency-based Semi-supervised Learning for Object Detection(leverage unlabled data, consistency loss)

Panoptic Segmentation

1. Panoptic Segmentation
2. Panoptic Feature Pyramid Networks
3. Fast Panoptic Segmentation Network👀
4. AdaptIS: Adaptive Instance Selection Network(AdaptIS)
5. SpatialFlow: Bridging All Tasks for Panoptic Segmentation(SpatialFlow)

Label Assignment

1. Learning from Noisy Anchors for One-stage Object Detection
2. Bridging the Gap Between Anchor-based and Anchor-free Detection via Adaptive Training Sample Selection(ATSS)
3. Libra R-CNN: Towards Balanced Learning for Object Detection(Libra RCNN)
4. IoU-uniform R-CNN- Breaking Through the Limitations of RPN(IoU-Uniform RCNN, deeply water paper)

NMS Improvement

1. Improving Object Detection With One Line of Code(SoftNMS)
2. Bounding Box Regression with Uncertainty for Accurate Object Detection(Ensemble of Bbox NMS)
3. Relation Networks for Object Detection👀
4. Learning Non-maximum Suppression👀
5. Acquisition of Localization Confidence for Accurate Object Detection(IoUNet, IoU-Guided NMS)
6. NMS by Representative Region: Towards Crowded Pedestrian Detection by Proposal Pairing
7. Detection in Crowded Scenes: One Proposal, Multiple Predictions

Object Detection

1. RDSNet: A New Deep Architecture for Reciprocal Object Detection and Instance Segmentation(RDSNet)
2. Empirical Upper-bound in Object Detection and More
3. Is Sampling Heuristics Necessary in Training Deep Object Detectors?(Bias Initialization, train RetinaNet w/o focal loss)
4. Deformable Convolutional Networks(self learnt offset for convolution)
5. Side-Aware Boundary Localization for More Precise Object Detection(SABL, slices for localization)
6. Multiple Anchor Learning for Visual Object Detection(iteratively optimize N->1 anchors with MIL)

Graph

1. DeepGCNs: Can GCNs Go as Deep as CNNs?

Pedestrian Detection

1. Mask-Guided Attention Network for Occluded Pedestrian Detection
2. Detection in Crowded Scenes: One Proposal, Multiple Predictions
3. NMS by Representative Region: Towards Crowded Pedestrian Detection by Proposal Pairing

To be read

CVPR2020

1. AugFPN: Improving Multi-scale Feature Learning for Object Detection
2. DR Loss: Improving Object Detection by Distributional Ranking
3. SpineNet: Learning Scale-Permuted Backbone for Recognition and Localization
4. D2Det: Towards High Quality Object Detection and Instance Segmentation
5. Overcoming Classifier Imbalance for Long-tail Object Detection with Balanced Group Softmax
6. Large-Scale Object Detection in the Wild from Imbalanced Multi-Labels
7. Seeing without Looking: Contextual Rescoring of Object Detections for AP Maximization
8. Learning a Unified Sample Weighting Network for Object Detection
9. BlendMask: Top-Down Meets Bottom-Up for Instance Segmentation
10. CenterMask : Real-Time Anchor-Free Instance Segmentation
11. Deep Snake for Real-Time Instance Segmentation
12. Offset Bin Classification Network for Accurate Object Detection