Paper Reading
-
이번에는 CVPR 2023에 게재된 논문인 ocus On Details: Online Multi-object Tracking with Diverse Fine-grained Representation를 읽고, 리뷰해보고자 합니다.Index1. Background 1.1. Triplet Loss2. Abstract3. Introduction4. Related Work5. Method 5.1. Overview 5.2. Flow Aligned FPN 5.3. Multi-head Part Mask Generator 5.4. Train and Inference6. Experiment7. Conclusion 1. Background1.1. Triplet Loss Triplet Loss1. ..
[리뷰] Focus On Details: Online Multi-object Tracking with Diverse Fine-grained Representation이번에는 CVPR 2023에 게재된 논문인 ocus On Details: Online Multi-object Tracking with Diverse Fine-grained Representation를 읽고, 리뷰해보고자 합니다.Index1. Background 1.1. Triplet Loss2. Abstract3. Introduction4. Related Work5. Method 5.1. Overview 5.2. Flow Aligned FPN 5.3. Multi-head Part Mask Generator 5.4. Train and Inference6. Experiment7. Conclusion 1. Background1.1. Triplet Loss Triplet Loss1. ..
2024.12.31 -
이번에는 한국컴퓨터정보학회 2022 동계학술대회 논문집 제30권 1호에 발표된 논문인 Target Detection Method using Lightweight Mean Shift Segmentation and Shape Features를 읽고, 이전과 다르게 세세한 리뷰 보다는, 요약 및 논문에 대한 내 생각을 위주로 정리해보고자 합니다.Index1. Summary2. My Opinion1. Summary본 논문은 하드웨어 리소스가 제한된 장치에서 영상 분할 방법을 이용하여 small size object의detection을 하는 방법의 개선 방안에 대해 제안한다.영상 분할 방법은 region 기반 방법과, edge 기반 방법으로 나누어지며 대표적으로 FCM, Quickshift, Felzenszwal..
[리뷰] Target Detection Method using Lightweight Mean Shift Segmentation and Shape Features이번에는 한국컴퓨터정보학회 2022 동계학술대회 논문집 제30권 1호에 발표된 논문인 Target Detection Method using Lightweight Mean Shift Segmentation and Shape Features를 읽고, 이전과 다르게 세세한 리뷰 보다는, 요약 및 논문에 대한 내 생각을 위주로 정리해보고자 합니다.Index1. Summary2. My Opinion1. Summary본 논문은 하드웨어 리소스가 제한된 장치에서 영상 분할 방법을 이용하여 small size object의detection을 하는 방법의 개선 방안에 대해 제안한다.영상 분할 방법은 region 기반 방법과, edge 기반 방법으로 나누어지며 대표적으로 FCM, Quickshift, Felzenszwal..
2024.09.10 -
이번에는 VisDrone 2021 ICCV workshop에 발표된 논문인 TPH-YOLOv5: Improved YOLOv5 Based on Transformer Prediction Head for Object Detection on Drone-captured Scenarios를 읽고, 리뷰해보고자 합니다. Index 1. Background 1.1. YOLO 1.2. Transformer 1.3. Convolutional Block Attention Model 2. Abstract 3. Introduction 4. Related Work 5. Method 5.1. Overview of YOLOv5 5.2. TPH-YOLOv5 6. Experiment 7. Conclusion 1. Background 1..
[리뷰] TPH-YOLOv5: Improved YOLOv5 Based on Transformer Prediction Head for Object Detection on Drone-captured Scenarios이번에는 VisDrone 2021 ICCV workshop에 발표된 논문인 TPH-YOLOv5: Improved YOLOv5 Based on Transformer Prediction Head for Object Detection on Drone-captured Scenarios를 읽고, 리뷰해보고자 합니다. Index 1. Background 1.1. YOLO 1.2. Transformer 1.3. Convolutional Block Attention Model 2. Abstract 3. Introduction 4. Related Work 5. Method 5.1. Overview of YOLOv5 5.2. TPH-YOLOv5 6. Experiment 7. Conclusion 1. Background 1..
2023.11.14 -
이번에는 ISRPS 2022에 발표된 논문인 A Normalized Gaussian Wasserstein Distance for Tiny Object Detection를 읽고, 이전과 다르게 세세한 리뷰 보다는, 새롭게 제시된 개념 위주로 간략하게 훑어보고자 합니다. Index 1. Background 1.1. Problem about tiny object detection 1.2. 다양한 IoU의 종류 2. Method 2.1. Wasserstein Distance 2.2. Gaussian Wassertein Distance 2.3. Normalized Gaussian Wassertein Distance 3. Experiment 1. Background 1.1. Problem about tiny obj..
[리뷰] A Normalized Gaussian Wasserstein Distance for Tiny Object Detection이번에는 ISRPS 2022에 발표된 논문인 A Normalized Gaussian Wasserstein Distance for Tiny Object Detection를 읽고, 이전과 다르게 세세한 리뷰 보다는, 새롭게 제시된 개념 위주로 간략하게 훑어보고자 합니다. Index 1. Background 1.1. Problem about tiny object detection 1.2. 다양한 IoU의 종류 2. Method 2.1. Wasserstein Distance 2.2. Gaussian Wassertein Distance 2.3. Normalized Gaussian Wassertein Distance 3. Experiment 1. Background 1.1. Problem about tiny obj..
2023.09.07 -
이번에는 NeurIPS 2022에 발표된 논문인 Quo Vadis: Is Trajectory Forecasting the Key Towards Long-Term Multi-Object Tracking?를 읽고, 리뷰해보고자 합니다. Index 1. Background 1.1. Homography 1.2. Graph Neural Network 1.3. Bi-Partite Matching 2. Abstract 3. Introduction 4. Related Work 5. Method 5.1. Overview 5.2. Data-driven Homography Estimation 5.3. Forecasting 5.4. Tracking via Forecasting 6. Experiment 7. Conclusion..
[리뷰] Quo Vadis: Is Trajectory Forecasting the Key Towards Long-Term Multi-Object Tracking?이번에는 NeurIPS 2022에 발표된 논문인 Quo Vadis: Is Trajectory Forecasting the Key Towards Long-Term Multi-Object Tracking?를 읽고, 리뷰해보고자 합니다. Index 1. Background 1.1. Homography 1.2. Graph Neural Network 1.3. Bi-Partite Matching 2. Abstract 3. Introduction 4. Related Work 5. Method 5.1. Overview 5.2. Data-driven Homography Estimation 5.3. Forecasting 5.4. Tracking via Forecasting 6. Experiment 7. Conclusion..
2023.04.19 -
이번에는 IEEE Transactions on Signal Processing 2022에 발표된 논문인 KalmanNet: Neural Network Aided Kalman Filtering for Partially Known Dynamics를 읽고, 리뷰해보고자 합니다.Index1. Background 1.1. State Space Model 1.2. Data-Aided Filtering Problem Formulation 1.3. Extended Kalman Filter 1.4. Recurrent Neural Network 1.5. Back Propagation Through Time 1.6. Truncated BPTT 1.7. Gated Recurrent Uni..
[리뷰] KalmanNet: Neural Network Aided Kalman Filtering for Partially Known Dynamics이번에는 IEEE Transactions on Signal Processing 2022에 발표된 논문인 KalmanNet: Neural Network Aided Kalman Filtering for Partially Known Dynamics를 읽고, 리뷰해보고자 합니다.Index1. Background 1.1. State Space Model 1.2. Data-Aided Filtering Problem Formulation 1.3. Extended Kalman Filter 1.4. Recurrent Neural Network 1.5. Back Propagation Through Time 1.6. Truncated BPTT 1.7. Gated Recurrent Uni..
2023.03.27 -
이번에는 WACV 2023에 발표된 논문인 Hard to Track Object with Irregular Motions and Similar Appearances? Make It Easier by Buffering the Matching Space를 읽고, 리뷰해보고자 합니다. Index 1. Background 1.1. Generalized IoU 1.2. Distance IoU 1.3. Complete IoU 2. Abstract 3. Introduction 4. Related Work 5. Method 5.1. Tracking Pipeline 5.2. Buffered IoU 5.3. Simple Motion Estimation 5.4. Track Management 6. Experiment 7. ..
[리뷰] Hard to Track Object with Irregular Motions and Similar Appearances? Make It Easier by Buffering the Matching Space이번에는 WACV 2023에 발표된 논문인 Hard to Track Object with Irregular Motions and Similar Appearances? Make It Easier by Buffering the Matching Space를 읽고, 리뷰해보고자 합니다. Index 1. Background 1.1. Generalized IoU 1.2. Distance IoU 1.3. Complete IoU 2. Abstract 3. Introduction 4. Related Work 5. Method 5.1. Tracking Pipeline 5.2. Buffered IoU 5.3. Simple Motion Estimation 5.4. Track Management 6. Experiment 7. ..
2023.03.21 -
이번에는 2022년에 발표된 논문인 Observation-Centric SORT: Rethinking SORT for Robust Multi-Object Tracking를 읽고, 리뷰해보고자 합니다. 본 논문은 2023년에 발표된 논문인 Deep OC-SORT: Multi-Pedestrian Tracking by Adaptive Re-Identification의 baseline이 되는 논문입니다. Index 1. Background 1.1. Limitations of Simple Online and Realtime Tracking 2. Abstract 3. Introduction 4. Related Work 5. Method 5.1. Observation-centric Online Smoothing 5...
[리뷰] Observation-Centric SORT: Rethinking SORT for Robust Multi-Object Tracking이번에는 2022년에 발표된 논문인 Observation-Centric SORT: Rethinking SORT for Robust Multi-Object Tracking를 읽고, 리뷰해보고자 합니다. 본 논문은 2023년에 발표된 논문인 Deep OC-SORT: Multi-Pedestrian Tracking by Adaptive Re-Identification의 baseline이 되는 논문입니다. Index 1. Background 1.1. Limitations of Simple Online and Realtime Tracking 2. Abstract 3. Introduction 4. Related Work 5. Method 5.1. Observation-centric Online Smoothing 5...
2023.03.13 -
이번에는 2022년에 발표된 논문인 SMILEtrack: SiMIlarity LEarning for Multiple Object Tracking를 읽고, 리뷰해보고자 합니다. Index 1. Background 1.1. Seperate Detection and Embedding 1.2. ByteTrack 2. Abstract 3. Introduction 3.1. Previous Method 3.2. Proposed Method 4. Related Work 5. Method 5.1. Architecture Overview 5.2. Similarity Learning Module for Re-ID 5.3. The Image Slicing Attention Block 5.4. Image Slicing 5.5...
[리뷰] SMILEtrack: SiMIlarity LEarning for Multiple Object Tracking이번에는 2022년에 발표된 논문인 SMILEtrack: SiMIlarity LEarning for Multiple Object Tracking를 읽고, 리뷰해보고자 합니다. Index 1. Background 1.1. Seperate Detection and Embedding 1.2. ByteTrack 2. Abstract 3. Introduction 3.1. Previous Method 3.2. Proposed Method 4. Related Work 5. Method 5.1. Architecture Overview 5.2. Similarity Learning Module for Re-ID 5.3. The Image Slicing Attention Block 5.4. Image Slicing 5.5...
2023.03.06 -
이번에는 2022년에 발표된 논문인 BoT-SORT: Robust Associations Multi-Pedestrian Tracking를 읽고, 리뷰해보고자 합니다. Index 1. Background 1.1. Extrapolaltion 1.2. Linear Kalman Filter 1.3. RANdom SAmple Consensus 1.4. Rigid Motion & Non-Rigid Motion 2. Abstract 3. Introduction 4. Related Work 5. Method 5.1. Kalman Filter 5.2. Camera Motion Compensation 5.3. IoU - Re-ID Fusion 5.4. Whole Architecture 6. Experiment 7. Con..
[리뷰] BoT-SORT: Robust Associations Multi-Pedestrian Tracking이번에는 2022년에 발표된 논문인 BoT-SORT: Robust Associations Multi-Pedestrian Tracking를 읽고, 리뷰해보고자 합니다. Index 1. Background 1.1. Extrapolaltion 1.2. Linear Kalman Filter 1.3. RANdom SAmple Consensus 1.4. Rigid Motion & Non-Rigid Motion 2. Abstract 3. Introduction 4. Related Work 5. Method 5.1. Kalman Filter 5.2. Camera Motion Compensation 5.3. IoU - Re-ID Fusion 5.4. Whole Architecture 6. Experiment 7. Con..
2023.02.27 -
이번에는 IEEE 2023에 게재된 논문인 StrongSORT: Make DeepSORT Great Again를 읽고, 리뷰해보고자 합니다. Index 1. Background 1.1. Mahalanobis Distance 1.2. Exponential Moving Average 1.3. High Order Tracking Accuracy 2. Abstract 3. Introduction 4. Related Work 5. Method 5.1. Review of DeepSORT 5.2. Stronger DeepSORT 5.3. Appearance-Free Link model 5.4. Gaussian-Smoothed Interpolation 6. Experiment 7. Conclusion 1. Backgr..
[리뷰] StrongSORT: Make DeepSORT Great Again이번에는 IEEE 2023에 게재된 논문인 StrongSORT: Make DeepSORT Great Again를 읽고, 리뷰해보고자 합니다. Index 1. Background 1.1. Mahalanobis Distance 1.2. Exponential Moving Average 1.3. High Order Tracking Accuracy 2. Abstract 3. Introduction 4. Related Work 5. Method 5.1. Review of DeepSORT 5.2. Stronger DeepSORT 5.3. Appearance-Free Link model 5.4. Gaussian-Smoothed Interpolation 6. Experiment 7. Conclusion 1. Backgr..
2023.02.24 -
이번에는 ECCV 2022에 게재된 논문인 MOTR: End-to-End Multiple-Object Tracking with TRansformer를 읽고, 리뷰해보고자 합니다. Index 1. Background 1.1. Tracking by Detection 1.2. Object Query 2. Abstract 3. Introduction 4. Related Work 5. Method 5.1. Revisiting Deformable DETR 5.2. Whole Architecture 5.3. Track Query 5.4. Continuous Query Passing 5.5. Query Interaction 5.5.1. Query Interaction Module 5.5.2. Temporal Aggre..
[리뷰] MOTR: End-to-End Multiple-Object Tracking with TRansformer이번에는 ECCV 2022에 게재된 논문인 MOTR: End-to-End Multiple-Object Tracking with TRansformer를 읽고, 리뷰해보고자 합니다. Index 1. Background 1.1. Tracking by Detection 1.2. Object Query 2. Abstract 3. Introduction 4. Related Work 5. Method 5.1. Revisiting Deformable DETR 5.2. Whole Architecture 5.3. Track Query 5.4. Continuous Query Passing 5.5. Query Interaction 5.5.1. Query Interaction Module 5.5.2. Temporal Aggre..
2023.02.01