{"url":"/method/atss","slug":"atss","name":"ATSS","full_name":"Adaptive Training Sample Selection","full_name_withheld":false,"description_markdown":"**Adaptive Training Sample Selection**, or **ATSS**, is a method to automatically select positive and negative samples according to statistical characteristics of object. It bridges the gap between anchor-based and anchor-free detectors. \r\n\r\nFor each ground-truth box $g$ on the image, we first find out its candidate positive samples. As described in Line $3$ to $6$, on each pyramid level, we select $k$ anchor boxes whose center are closest to the center of $g$ based on L2 distance. Supposing there are $\\mathcal{L}$ feature pyramid levels, the ground-truth box $g$ will have $k\\times\\mathcal{L}$ candidate positive samples. After that, we compute the IoU between these candidates and the ground-truth $g$ as $\\mathcal{D}_g$ in Line $7$, whose mean and standard deviation are computed as $m_g$ and $v_g$ in Line $8$ and Line $9$. With these statistics, the IoU threshold for this ground-truth $g$ is obtained as $t_g=m_g+v_g$ in Line $10$. Finally, we select these candidates whose IoU are greater than or equal to the threshold $t_g$ as final positive samples in Line $11$ to $15$. \r\n\r\nNotably ATSS also limits the positive samples' center to the ground-truth box as shown in Line $12$. Besides, if an anchor box is assigned to multiple ground-truth boxes, the one with the highest IoU will be selected. The rest are negative samples.","description_state":"present","introduced_year":null,"introduced_by":{"title":null,"paper":null,"first_author":null,"n_authors":0,"url_abs":null,"archive_paper_url":null},"source":{"url":"https://arxiv.org/abs/1912.02424v4","title":"Bridging the Gap Between Anchor-based and Anchor-free Detection via Adaptive Training Sample Selection","url_on_a_paper_host":true},"code_snippet_url":"https://github.com/sfzhang15/ATSS/blob/79dfb28bd18c931dd75a3ca2c63d32f5e4b1626a/atss_core/modeling/rpn/atss/atss.py#L194","code_snippet_url_on_a_code_host":true,"categories":[{"area":"General","area_id":"general","collection":"Prioritized Sampling","url":"/methods/category/prioritized-sampling","pwc_aliases":[]}],"n_papers_tagged":16,"archive_num_papers":null,"papers_newest_first":[{"paper":"/paper/2408-02561","title":"HQOD: Harmonious Quantization for Object Detection","date":"2024-08-05","arxiv_id":"2408.02561","n_code_links":1,"syntology":null},{"paper":null,"title":"Generalized Mask-aware IoU for Anchor Assignment for Real-time Instance Segmentation","date":"2023-12-28","arxiv_id":"2312.17031","n_code_links":0,"syntology":null},{"paper":"/paper/being-aware-of-localization-accuracy-by","title":"Being Aware of Localization Accuracy By Generating Predicted-IoU-Guided Quality Scores","date":"2023-09-23","arxiv_id":"2309.13269","n_code_links":1,"syntology":null},{"paper":"/paper/how-to-train-an-accurate-and-efficient-object","title":"How to Train an Accurate and Efficient Object Detection Model on Any Dataset","date":"2022-11-30","arxiv_id":"2211.17170","n_code_links":1,"syntology":null},{"paper":"/paper/detrs-with-collaborative-hybrid-assignments","title":"DETRs with Collaborative Hybrid Assignments Training","date":"2022-11-22","arxiv_id":"2211.12860","n_code_links":6,"syntology":{"ran":0,"of":5,"unverified":5,"pointer_only":0}},{"paper":null,"title":"Sample hardness based gradient loss for long-tailed cervical cell detection","date":"2022-08-07","arxiv_id":"2208.03779","n_code_links":0,"syntology":null},{"paper":"/paper/recursivemix-mixed-learning-with-history","title":"RecursiveMix: Mixed Learning with History","date":"2022-03-14","arxiv_id":"2203.06844","n_code_links":1,"syntology":null},{"paper":"/paper/mask-aware-iou-for-anchor-assignment-in-real","title":"Mask-aware IoU for Anchor Assignment in Real-time Instance Segmentation","date":"2021-10-19","arxiv_id":"2110.09734","n_code_links":1,"syntology":null},{"paper":"/paper/tood-task-aligned-one-stage-object-detection","title":"TOOD: Task-aligned One-stage Object Detection","date":"2021-08-17","arxiv_id":"2108.07755","n_code_links":6,"syntology":{"ran":2,"of":4,"unverified":2,"pointer_only":2}},{"paper":"/paper/disentangle-your-dense-object-detector","title":"Disentangle Your Dense Object Detector","date":"2021-07-07","arxiv_id":"2107.02963","n_code_links":2,"syntology":null},{"paper":"/paper/control-distance-iou-and-control-distance-iou","title":"Control Distance IoU and Control Distance IoU Loss Function for Better Bounding Box Regression","date":"2021-03-22","arxiv_id":"2103.11696","n_code_links":1,"syntology":null},{"paper":null,"title":"Implicit Feature Pyramid Network for Object Detection","date":"2020-12-25","arxiv_id":"2012.13563","n_code_links":0,"syntology":null},{"paper":"/paper/generalized-focal-loss-v2-learning-reliable","title":"Generalized Focal Loss V2: Learning Reliable Localization Quality Estimation for Dense Object Detection","date":"2020-11-25","arxiv_id":"2011.12885","n_code_links":5,"syntology":null},{"paper":"/paper/relationnet-bridging-visual-representations","title":"RelationNet++: Bridging Visual Representations for Object Detection via Transformer Decoder","date":"2020-10-29","arxiv_id":"2010.15831","n_code_links":4,"syntology":{"ran":0,"of":5,"unverified":5,"pointer_only":0}},{"paper":"/paper/generalized-focal-loss-learning-qualified-and","title":"Generalized Focal Loss: Learning Qualified and Distributed Bounding Boxes for Dense Object Detection","date":"2020-06-08","arxiv_id":"2006.04388","n_code_links":7,"syntology":{"ran":1,"of":24,"unverified":23,"pointer_only":0}},{"paper":"/paper/bridging-the-gap-between-anchor-based-and","title":"Bridging the Gap Between Anchor-based and Anchor-free Detection via Adaptive Training Sample Selection","date":"2019-12-05","arxiv_id":"1912.02424","n_code_links":13,"syntology":{"ran":0,"of":4,"unverified":4,"pointer_only":0}}],"papers_shown":16,"tasks":[{"task":"/task/object-detection","name":"Object Detection","papers":12},{"task":"/task/object-detection-1","name":"object-detection","papers":10},{"task":"/task/object","name":"Object","papers":6},{"task":"/task/instance-segmentation","name":"Instance Segmentation","papers":3},{"task":"/task/semantic-segmentation","name":"Semantic Segmentation","papers":3},{"task":"/task/regression-1","name":"regression","papers":3},{"task":"/task/decoder","name":"Decoder","papers":2},{"task":"/task/dense-object-detection","name":"Dense Object Detection","papers":2},{"task":"/task/real-time-instance-segmentation","name":"Real-time Instance Segmentation","papers":2},{"task":"/task/segmentation","name":"Segmentation","papers":2},{"task":"/task/2d-object-detection","name":"2D Object Detection","papers":1},{"task":null,"name":"CPU","papers":1},{"task":"/task/cell-detection","name":"Cell Detection","papers":1},{"task":"/task/classification-1","name":"Classification","papers":1},{"task":"/task/disentanglement","name":"Disentanglement","papers":1},{"task":null,"name":"GPU","papers":1},{"task":"/task/classification","name":"General Classification","papers":1},{"task":"/task/quantization","name":"Quantization","papers":1},{"task":"/task/video-understanding","name":"Video Understanding","papers":1},{"task":"/task/set-matching","name":"set matching","papers":1}],"tasks_shown":20,"n_tasks":20,"usage_by_year":[{"year":"2019","papers":1},{"year":"2020","papers":4},{"year":"2021","papers":4},{"year":"2022","papers":4},{"year":"2023","papers":2},{"year":"2024","papers":1}],"row_source":"embedded","archive":{"source":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","archive_url":"https://paperswithcode.com/method/atss"},"syntology_read_at":"2026-09-24T18:15:14+00:00"}