{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/dsod-learning-deeply-supervised-object","title":"DSOD: Learning Deeply Supervised Object Detectors from Scratch","arxiv_id":"1708.01241","date":"2017-08-03","proceeding":"ICCV 2017 10","authors":["Zhiqiang Shen","Zhuang Liu","Jianguo Li","Yu-Gang Jiang","Yurong Chen","xiangyang xue"],"abstract":"We present Deeply Supervised Object Detector (DSOD), a framework that can\nlearn object detectors from scratch. State-of-the-art object objectors rely\nheavily on the off-the-shelf networks pre-trained on large-scale classification\ndatasets like ImageNet, which incurs learning bias due to the difference on\nboth the loss functions and the category distributions between classification\nand detection tasks. Model fine-tuning for the detection task could alleviate\nthis bias to some extent but not fundamentally. Besides, transferring\npre-trained models from classification to detection between discrepant domains\nis even more difficult (e.g. RGB to depth images). A better solution to tackle\nthese two critical problems is to train object detectors from scratch, which\nmotivates our proposed DSOD. Previous efforts in this direction mostly failed\ndue to much more complicated loss functions and limited training data in object\ndetection. In DSOD, we contribute a set of design principles for training\nobject detectors from scratch. One of the key findings is that deep\nsupervision, enabled by dense layer-wise connections, plays a critical role in\nlearning a good detector. Combining with several other principles, we develop\nDSOD following the single-shot detection (SSD) framework. Experiments on PASCAL\nVOC 2007, 2012 and MS COCO datasets demonstrate that DSOD can achieve better\nresults than the state-of-the-art solutions with much more compact models. For\ninstance, DSOD outperforms SSD on all three benchmarks with real-time detection\nspeed, while requires only 1/2 parameters to SSD and 1/10 parameters to Faster\nRCNN. Our code and models are available at: https://github.com/szq0214/DSOD .","url_abs":"http://arxiv.org/abs/1708.01241v2","url_pdf":"http://arxiv.org/pdf/1708.01241v2.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"dsod-learning-deeply-supervised-object","repo_url":"https://github.com/szq0214/DSOD","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"caffe2","reach":{"status":"ok","spdx":"NOASSERTION"}},{"paper_slug":"dsod-learning-deeply-supervised-object","repo_url":"https://github.com/CeLaMbDa/DSOD_keras","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":null},{"paper_slug":"dsod-learning-deeply-supervised-object","repo_url":"https://github.com/szq0214/GFR-DSOD","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null},{"paper_slug":"dsod-learning-deeply-supervised-object","repo_url":"https://github.com/szq0214/GRP-DSOD","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"classification","task_name":"General Classification"},{"task_slug":"object","task_name":"Object"},{"task_slug":"object-detection","task_name":"Object Detection"},{"task_slug":"object-detection-1","task_name":"object-detection"}],"methods":[{"method_slug":"1x1-convolution","method_name":"1x1 Convolution"},{"method_slug":"convolution","method_name":"Convolution"},{"method_slug":"non-maximum-suppression","method_name":"Non Maximum Suppression"},{"method_slug":"ssd","method_name":"SSD"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1708.01241","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}