{"id":26256798,"url":"https://github.com/mahshid1378/mahshiddetr","last_synced_at":"2026-05-03T18:34:56.891Z","repository":{"id":279220384,"uuid":"938083843","full_name":"mahshid1378/MAHSHIDDETR","owner":"mahshid1378","description":"DETR++: Official PyTorch Implementation","archived":false,"fork":false,"pushed_at":"2025-02-24T12:26:05.000Z","size":1609,"stargazers_count":0,"open_issues_count":0,"forks_count":0,"subscribers_count":1,"default_branch":"main","last_synced_at":"2025-12-28T06:11:23.787Z","etag":null,"topics":["computer-vision","cvpr","cvpr2022","deeplearning","detection","detr","machine-learning","object-detection","pythorch","transformer","vision","vision-transformer"],"latest_commit_sha":null,"homepage":"","language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"mit","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/mahshid1378.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null}},"created_at":"2025-02-24T11:52:30.000Z","updated_at":"2025-02-24T12:26:08.000Z","dependencies_parsed_at":"2025-02-24T13:24:09.298Z","dependency_job_id":"08219144-738a-463d-b080-651632949d17","html_url":"https://github.com/mahshid1378/MAHSHIDDETR","commit_stats":null,"previous_names":["mahshid1378/mahshiddetr"],"tags_count":0,"template":false,"template_full_name":null,"purl":"pkg:github/mahshid1378/MAHSHIDDETR","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/mahshid1378%2FMAHSHIDDETR","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/mahshid1378%2FMAHSHIDDETR/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/mahshid1378%2FMAHSHIDDETR/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/mahshid1378%2FMAHSHIDDETR/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/mahshid1378","download_url":"https://codeload.github.com/mahshid1378/MAHSHIDDETR/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/mahshid1378%2FMAHSHIDDETR/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":32580013,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-05-03T06:36:36.687Z","status":"ssl_error","status_checked_at":"2026-05-03T06:36:09.306Z","response_time":103,"last_error":"SSL_read: unexpected eof while reading","robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":false,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["computer-vision","cvpr","cvpr2022","deeplearning","detection","detr","machine-learning","object-detection","pythorch","transformer","vision","vision-transformer"],"created_at":"2025-03-13T20:19:28.112Z","updated_at":"2026-05-03T18:34:56.848Z","avatar_url":"https://github.com/mahshid1378.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"# SAM-DETR (Semantic-Aligned-Matching DETR)\n\n[![arXiv](https://img.shields.io/badge/arXiv-2203.06883-b31b1b.svg)](https://arxiv.org/abs/2203.06883)\n[![Survey](https://github.com/sindresorhus/awesome/blob/main/media/mentioned-badge.svg)](https://github.com/dk-liang/Awesome-Visual-Transformer) \n[![Maintenance](https://img.shields.io/badge/Maintained%3F-yes-green.svg)](https://GitHub.com/Naereen/StrapDown.js/graphs/commit-activity) \n[![PR's Welcome](https://img.shields.io/badge/PRs-welcome-brightgreen.svg?style=flat)](http://makeapullrequest.com) \n[![GitHub license](https://badgen.net/github/license/ZhangGongjie/SAM-DETR)](https://github.com/ZhangGongjie/SAM-DETR/blob/master/LICENSE)\n\nThis repository is an official PyTorch implementation of the\nCVPR 2022 paper \"[Accelerating DETR Convergence via Semantic-Aligned Matching](https://arxiv.org/abs/2203.06883)\". \n\n\u003cb\u003e*[UPDATE on 21 Apr 2022]*\u003c/b\u003e \u0026nbsp;  We found that with a very simple modification (with no extra computational cost), SAM-DETR can achieve better performance. On MS-COCO, **SAM-DETR w/ SMCA** can achieve **37.0 AP** within 12 epochs, and **42.7 AP** within 50 epochs. We will release the updated training scripts, model weights, and logs in the future. Please stay tuned!\n\n## Introduction\n\n\u003cb\u003e TL;DR \u003c/b\u003e \u0026nbsp; SAM-DETR is an efficeint DETR-like object detector that can\nconverge wihtin 12 epochs and outperform the strong Faster R-CNN (w/ FPN) baseline.\n\nThe recently developed DEtection TRansformer (DETR) has established a new\nobject detection paradigm by eliminating a series of hand-crafted components.\nHowever, DETR suffers from extremely slow convergence, which increases the\ntraining cost significantly. We observe that the slow convergence can be largely\nattributed to the complication in matching object queries to encoded image features\nin DETR's decoder cross-attention modules.\n\n\u003cdiv align=center\u003e  \n\u003cimg src='.assets/matching_complication.jpg' width=\"70%\"\u003e\n\u003c/div\u003e\n\nMotivated by this observation, in our paper, we propose SAM-DETR, a\nSemantic-Aligned-Matching DETR that can greatly accelerates DETR's convergence\nwithout sacrificing its accuracy. SAM-DETR addresses the slow convergence issue\nfrom two perspectives. First, it projects object queries into the same\nembedding space as encoded image features, where the matching can be accomplished\nefficiently with aligned semantics. Second, it explicitly searches salient\npoints with the most discriminative features for semantic-aligned matching,\nwhich further speeds up the convergence and boosts detection accuracy as well.\nBeing like a plug and play, SAM-DETR complements existing convergence solutions\nwell yet only introduces slight computational overhead. Experiments\nshow that the proposed SAM-DETR achieves superior convergence as well as\ncompetitive detection accuracy.\n\nAt the core of SAM-DETR is a plug-and-play module named \"Semantics Aligner\" appended\nahead of the cross-attention module in DETR's each decoder layer. It also models a learnable\nreference box for each object query, whose center location is used to generate\ncorresponding position embeddings.\n\n\u003cdiv align=center\u003e  \n\u003cimg src='.assets/decoder_layer.jpg' width=\"90%\"\u003e\n\u003c/div\u003e\n\nThe figure below illustrates the architecture of the appended \"Semantics Aligner\", which\naligns the semantics of \"encoded image features\" and \"object queries\" by re-sampling features \nfrom multiple salient points as new object queries.\n\n\u003cdiv align=center\u003e  \n\u003cimg src='.assets/semantics_aligner.jpg' width=\"78%\"\u003e\n\u003c/div\u003e\n\nBeing like a plug-and-play, our approach can be\neasily integrated with existing convergence solutions (*e.g.*, SMCA) in a complementary manner,\nboosting detection accuracy and convergence speed further.\n\nPlease check [our CVPR 2022 paper](https://arxiv.org/abs/2203.06883) for more details.\n\n\n\n\n\n\n\n## Installation\n\n### Pre-Requisites\nYou must have NVIDIA GPUs to run the codes.\n\nThe implementation codes are developed and tested with the following environment setups:\n- Linux\n- 8x NVIDIA V100 GPUs (32GB)\n- CUDA 10.1\n- Python == 3.8\n- PyTorch == 1.8.1+cu101, TorchVision == 0.9.1+cu101\n- GCC == 7.5.0\n- cython, pycocotools, tqdm, scipy\n\nWe recommend using the exact setups above. However, other environments (Linux, Python\u003e=3.7, CUDA\u003e=9.2, GCC\u003e=5.4, PyTorch\u003e=1.5.1, TorchVision\u003e=0.6.1) should also work.\n\n### Code Installation\n\nFirst, clone the repository locally:\n```shell\ngit clone https://github.com/ZhangGongjie/SAM-DETR.git\n```\n\nWe recommend you to use [Anaconda](https://www.anaconda.com/) to create a conda environment:\n```bash\nconda create -n sam_detr python=3.8 pip\n```\n\nThen, activate the environment:\n```bash\nconda activate sam_detr\n```\n\nThen, install PyTorch and TorchVision:\n\n(preferably using our recommended setups; CUDA version should match your own local environment)\n```bash\nconda install pytorch=1.8.1 torchvision=0.9.1 cudatoolkit=10.1 -c pytorch\n```\n\nAfter that, install other requirements:\n```bash\nconda install cython scipy tqdm\npip install -U 'git+https://github.com/cocodataset/cocoapi.git#subdirectory=PythonAPI'\n```\n\n\u003cb\u003e*[Optional]*\u003c/b\u003e \u0026nbsp;  If you wish to run multi-scale version of SAM-DETR (results not reported in the CVPR paper), you need to compile [*Deformable Attention*],\nwhich is used in DETR encoder to generate feature pyramid efficiently. If you don't need multi-scale\nversion of SAM-DETR, you may skip this step.\n```bash\n# Optionally compile CUDA operators of Deformable Attention for multi-scale SAM-DETR\ncd SAM-DETR\ncd ./models/ops\nsh ./make.sh\npython test.py  # unit test (should see all checking is True)\n```\n\n### Data Preparation\n\nPlease download [COCO 2017 dataset](https://cocodataset.org/) and organize them as following:\n\n```\ncode_root/\n└── data/\n    └── coco/\n        ├── train2017/\n        ├── val2017/\n        └── annotations/\n        \t├── instances_train2017.json\n        \t└── instances_val2017.json\n```\n\n\n\n\n\n\n\n## Usage\n\n### Reproducing Paper Results\n\nAll scripts to reproduce results reported in [our CVPR 2022 paper](https://arxiv.org/abs/2203.06883)\nare stored in ```./scripts```. We also provide scripts for slurm cluster,\nwhich are stored in ```./scripts_slurm```.\n\nTaking \u003cb\u003eSAM-DETR-R50 w/ SMCA (12 epochs)\u003c/b\u003e for example, to reproduce its results, simply\nrun:\n```shell\nbash scripts/r50_smca_e12_4gpu.sh\n```\n\nTaking \u003cb\u003eSAM-DETR-R50 multiscale w/ SMCA (50 epochs)\u003c/b\u003e for example, to reproduce its results on a slurm cluster, simply\nrun:\n```shell\nbash scripts_slurm/r50_ms_smca_e50_8gpu.sh\n```\n\nReminder: To reproduce results, please make sure the total batch size matches the implementation details described in our paper. For ```R50 (single-scale)```\nexperiments, we use 4 GPUs with a batch size of 4 on each GPU. For ```R50 (multi-scale)```\nexperiments, we use 8 GPUs with a batch size of 2 on each GPU. For ```R50-DC5 (single-scale)```\nexperiments, we use 8 GPUs with a batch size of 1 on each GPU.\n\n\n\n### Training\nTo perform training on COCO *train2017*, modify the arguments based on the scripts below:\n```shell\npython -m torch.distributed.launch \\\n    --nproc_per_node=4 \\        # number of GPUs to perform training\n    --use_env main.py \\\n    --batch_size 4 \\            # batch_size on individual GPU (this is *NOT* total batch_size)\n    --smca \\                    # to integrate with SMCA, remove this line to disable SMCA\n    --dilation \\                # to enable DC5, remove this line to disable DC5\n    --multiscale \\              # to enable multi-scale, remove this line to disable multiscale\n    --epochs 50 \\               # total number of epochs to train\n    --lr_drop 40 \\              # when to drop learning rate\n    --output_dir output/xxxx    # where to store outputs, remove this line for not storing outputs\n```\nMore arguments and their explanations are available at ```main.py```.\n\n### Evaluation\nTo evaluate a model on COCO *val2017*, simply add ```--resume``` and ```--eval``` arguments to your training scripts:\n```shell\npython -m torch.distributed.launch \\\n    --nproc_per_node=4 \\\n    --use_env main.py \\\n    --batch_size 4 \\\n    --smca \\\n    --dilation \\                \n    --multiscale \\ \n    --epochs 50 \\\n    --lr_drop 40 \\ \n    --resume \u003cpath/to/checkpoint.pth\u003e \\   # trained model weights\n    --eval \\                              # this means that only evaluation will be performed\n    --output_dir output/xxxx   \n```\n\n\n### Visualize Detection Results\nWe provide `demo.py`, which is a minimal implementation that allows users to visualize model's detection predictions. It performs detection on images inside the `./images` folder, and stores detection visualizations in that folder. Taking \u003cb\u003eSAM-DETR-R50 w/ SMCA (50 epochs)\u003c/b\u003e for example, simply run:\n```shell\npython demo.py \\                       # do NOT use distributed mode\n    --smca \\\n    --epochs 50 \\                      # you need to set this correct. See models/fast_detr.py L50-79 for details.\n    --resume \u003cpath/to/checkpoint.pth\u003e  # trained model weights\n```\n\n\n\n## Model Zoo\n\n*Trained model weights are stored in Google Drive.*\n\nThe original DETR models trained for 500 epochs:\n\n\u003ctable\u003e\n  \u003cthead\u003e\n    \u003ctr style=\"text-align: right;\"\u003e\n      \u003cth\u003eMethod\u003c/th\u003e\n      \u003cth\u003eEpochs\u003c/th\u003e\n      \u003cth\u003eParams (M)\u003c/th\u003e\n      \u003cth\u003eGFLOPs\u003c/th\u003e\n      \u003cth\u003eAP\u003c/th\u003e\n      \u003cth\u003eURL\u003c/th\u003e\n    \u003c/tr\u003e\n  \u003c/thead\u003e\n  \u003ctbody\u003e\n    \u003ctr\u003e\n      \u003ctd\u003eDETR-R50\u003c/td\u003e\n      \u003ctd\u003e500\u003c/td\u003e\n      \u003ctd\u003e41\u003c/td\u003e\n      \u003ctd\u003e86\u003c/td\u003e\n      \u003ctd\u003e42.0\u003c/td\u003e\n      \u003ctd\u003e\u003ca href=\"https://dl.fbaipublicfiles.com/detr/logs/detr-r50_log.txt\"\u003elog\u003c/a\u003e\u003c/td\u003e\n    \u003c/tr\u003e\n    \u003ctr\u003e\n      \u003ctd\u003eDETR-R50-DC5\u003c/td\u003e\n      \u003ctd\u003e500\u003c/td\u003e\n      \u003ctd\u003e41\u003c/td\u003e\n      \u003ctd\u003e187\u003c/td\u003e\n      \u003ctd\u003e43.3\u003c/td\u003e\n      \u003ctd\u003e\u003ca href=\"https://dl.fbaipublicfiles.com/detr/logs/detr-r50-dc5_log.txt\"\u003elog\u003c/a\u003e\u003c/td\u003e\n    \u003c/tr\u003e\n  \u003c/tbody\u003e\n\u003c/table\u003e\n\n\nOur proposed SAM-DETR models (results reported in [our CVPR paper](https://arxiv.org/abs/2203.06883)):\n\u003ctable\u003e\n  \u003cthead\u003e\n    \u003ctr style=\"text-align: right;\"\u003e\n      \u003cth\u003eMethod\u003c/th\u003e\n      \u003cth\u003eEpochs\u003c/th\u003e\n      \u003cth\u003eParams (M)\u003c/th\u003e\n      \u003cth\u003eGFLOPs\u003c/th\u003e\n      \u003cth\u003eAP\u003c/th\u003e\n      \u003cth\u003eURL\u003c/th\u003e\n    \u003c/tr\u003e\n  \u003c/thead\u003e\n  \u003ctbody\u003e\n    \u003ctr\u003e\n      \u003ctd\u003eSAM-DETR-R50\u003c/td\u003e\n      \u003ctd\u003e12\u003c/td\u003e\n      \u003ctd\u003e58\u003c/td\u003e\n      \u003ctd\u003e100\u003c/td\u003e\n      \u003ctd\u003e33.1\u003c/td\u003e\n      \u003ctd\u003e\u003ca href=\"https://drive.google.com/file/d/1RXGs50nXkYBlYKFjI8nzt2ZDfLEVsLJZ/view?usp=sharing\"\u003emodel\u003c/a\u003e \u003cbr/\u003e \u003ca href=\".assets/output_logs/r50_e12.txt\"\u003elog\u003c/a\u003e\u003c/td\u003e\n    \u003c/tr\u003e\n    \u003ctr\u003e\n      \u003ctd\u003eSAM-DETR-R50 w/ SMCA\u003c/td\u003e\n      \u003ctd\u003e12\u003c/td\u003e\n      \u003ctd\u003e58\u003c/td\u003e\n      \u003ctd\u003e100\u003c/td\u003e\n      \u003ctd\u003e36.0\u003c/td\u003e\n      \u003ctd\u003e\u003ca href=\"https://drive.google.com/file/d/16ucFXgf0lrgzLArdmFzn3oOn3eGUrCU4/view?usp=sharing\"\u003emodel\u003c/a\u003e \u003cbr/\u003e \u003ca href=\".assets/output_logs/r50_smca_e12.txt\"\u003elog\u003c/a\u003e\u003c/td\u003e\n    \u003c/tr\u003e\n    \u003ctr\u003e\n      \u003ctd\u003eSAM-DETR-R50-DC5\u003c/td\u003e\n      \u003ctd\u003e12\u003c/td\u003e\n      \u003ctd\u003e58\u003c/td\u003e\n      \u003ctd\u003e210\u003c/td\u003e\n      \u003ctd\u003e38.3\u003c/td\u003e\n      \u003ctd\u003e\u003ca href=\"https://drive.google.com/file/d/18kZUCGLIQKerzcIlh_Hh87V3wUkrTwEu/view?usp=sharing\"\u003emodel\u003c/a\u003e \u003cbr/\u003e \u003ca href=\".assets/output_logs/r50_dc5_e12.txt\"\u003elog\u003c/a\u003e\u003c/td\u003e\n    \u003c/tr\u003e\n    \u003ctr\u003e\n      \u003ctd\u003eSAM-DETR-R50-DC5 w/ SMCA\u003c/td\u003e\n      \u003ctd\u003e12\u003c/td\u003e\n      \u003ctd\u003e58\u003c/td\u003e\n      \u003ctd\u003e210\u003c/td\u003e\n      \u003ctd\u003e40.6\u003c/td\u003e\n      \u003ctd\u003e\u003ca href=\"https://drive.google.com/file/d/1gDfRo1DEv43d6DUMF1n_Lzv2lkRv0huE/view?usp=sharing\"\u003emodel\u003c/a\u003e \u003cbr/\u003e \u003ca href=\".assets/output_logs/r50_dc5_smca_e12.txt\"\u003elog\u003c/a\u003e\u003c/td\u003e\n    \u003c/tr\u003e\n    \u003ctr\u003e\n      \u003ctd\u003eSAM-DETR-R50\u003c/td\u003e\n      \u003ctd\u003e50\u003c/td\u003e\n      \u003ctd\u003e58\u003c/td\u003e\n      \u003ctd\u003e100\u003c/td\u003e\n      \u003ctd\u003e39.8\u003c/td\u003e\n      \u003ctd\u003e\u003ca href=\"https://drive.google.com/file/d/11BZHGv2UAtqECX10MZ3VMS81RqGQM5Aq/view?usp=sharing\"\u003emodel\u003c/a\u003e \u003cbr/\u003e \u003ca href=\".assets/output_logs/r50_e50.txt\"\u003elog\u003c/a\u003e\u003c/td\u003e\n    \u003c/tr\u003e\n    \u003ctr\u003e\n      \u003ctd\u003eSAM-DETR-R50 w/ SMCA\u003c/td\u003e\n      \u003ctd\u003e50\u003c/td\u003e\n      \u003ctd\u003e58\u003c/td\u003e\n      \u003ctd\u003e100\u003c/td\u003e\n      \u003ctd\u003e41.8\u003c/td\u003e\n      \u003ctd\u003e\u003ca href=\"https://drive.google.com/file/d/1XHDkMsVB-zncVRsf21Z1-hPxJJCi7TGN/view?usp=sharing\"\u003emodel\u003c/a\u003e \u003cbr/\u003e \u003ca href=\".assets/output_logs/r50_smca_e50.txt\"\u003elog\u003c/a\u003e\u003c/td\u003e\n    \u003c/tr\u003e\n    \u003ctr\u003e\n      \u003ctd\u003eSAM-DETR-R50-DC5\u003c/td\u003e\n      \u003ctd\u003e50\u003c/td\u003e\n      \u003ctd\u003e58\u003c/td\u003e\n      \u003ctd\u003e210\u003c/td\u003e\n      \u003ctd\u003e43.3\u003c/td\u003e\n      \u003ctd\u003e\u003ca href=\"https://drive.google.com/file/d/1XYLr3yhqmmT3RWIJ_4SN1S-bO4BtWgcX/view?usp=sharing\"\u003emodel\u003c/a\u003e \u003cbr/\u003e \u003ca href=\".assets/output_logs/r50_dc5_e50.txt\"\u003elog\u003c/a\u003e\u003c/td\u003e\n    \u003c/tr\u003e\n    \u003ctr\u003e\n      \u003ctd\u003eSAM-DETR-R50-DC5 w/ SMCA\u003c/td\u003e\n      \u003ctd\u003e50\u003c/td\u003e\n      \u003ctd\u003e58\u003c/td\u003e\n      \u003ctd\u003e210\u003c/td\u003e\n      \u003ctd\u003e45.0\u003c/td\u003e\n      \u003ctd\u003e\u003ca href=\"https://drive.google.com/file/d/1q_iWQOZl1zXtaVD4vI4V9piwj2tkE8ek/view?usp=sharing\"\u003emodel\u003c/a\u003e \u003cbr/\u003e \u003ca href=\".assets/output_logs/r50_dc5_smca_e50.txt\"\u003elog\u003c/a\u003e\u003c/td\u003e\n    \u003c/tr\u003e\n  \u003c/tbody\u003e\n\u003c/table\u003e\n\n\n\n\n\nOur proposed multi-scale SAM-DETR models (results to appear in a journal extension):\n\u003ctable\u003e\n  \u003cthead\u003e\n    \u003ctr style=\"text-align: right;\"\u003e\n      \u003cth\u003eMethod\u003c/th\u003e\n      \u003cth\u003eEpochs\u003c/th\u003e\n      \u003cth\u003eParams (M)\u003c/th\u003e\n      \u003cth\u003eGFLOPs\u003c/th\u003e\n      \u003cth\u003eAP\u003c/th\u003e\n      \u003cth\u003eURL\u003c/th\u003e\n    \u003c/tr\u003e\n  \u003c/thead\u003e\n  \u003ctbody\u003e\n    \u003ctr\u003e\n      \u003ctd\u003eSAM-DETR-R50-MS\u003c/td\u003e\n      \u003ctd\u003e12\u003c/td\u003e\n      \u003ctd\u003e55\u003c/td\u003e\n      \u003ctd\u003e203\u003c/td\u003e\n      \u003ctd\u003e41.1\u003c/td\u003e\n      \u003ctd\u003e\u003ca href=\"https://drive.google.com/file/d/1Rdp7QngEaS0mxRQoxq7pGUM7LkZtgyzx/view?usp=sharing\"\u003emodel\u003c/a\u003e \u003cbr/\u003e \u003ca href=\".assets/output_logs/r50_ms_e12.txt\"\u003elog\u003c/a\u003e\u003c/td\u003e\n    \u003c/tr\u003e\n    \u003ctr\u003e\n      \u003ctd\u003eSAM-DETR-R50-MS w/ SMCA\u003c/td\u003e\n      \u003ctd\u003e12\u003c/td\u003e\n      \u003ctd\u003e55\u003c/td\u003e\n      \u003ctd\u003e203\u003c/td\u003e\n      \u003ctd\u003e42.8\u003c/td\u003e\n      \u003ctd\u003e\u003ca href=\"https://drive.google.com/file/d/12xJJA-8P1YPfW3wp3l2aMdkQX_mCAOR5/view?usp=sharing\"\u003emodel\u003c/a\u003e \u003cbr/\u003e \u003ca href=\".assets/output_logs/r50_ms_smca_e12.txt\"\u003elog\u003c/a\u003e\u003c/td\u003e\n    \u003c/tr\u003e\n    \u003ctr\u003e\n      \u003ctd\u003eSAM-DETR-R50-MS\u003c/td\u003e\n      \u003ctd\u003e50\u003c/td\u003e\n      \u003ctd\u003e55\u003c/td\u003e\n      \u003ctd\u003e203\u003c/td\u003e\n      \u003ctd\u003e46.1\u003c/td\u003e\n      \u003ctd\u003e\u003ca href=\"https://drive.google.com/file/d/1Fb6MFyldV-tJNfL7t1gpV7Yvk1LvhNvQ/view?usp=sharing\"\u003emodel\u003c/a\u003e \u003cbr/\u003e \u003ca href=\".assets/output_logs/r50_ms_e50.txt\"\u003elog\u003c/a\u003e\u003c/td\u003e\n    \u003c/tr\u003e\n    \u003ctr\u003e\n      \u003ctd\u003eSAM-DETR-R50-MS w/ SMCA\u003c/td\u003e\n      \u003ctd\u003e50\u003c/td\u003e\n      \u003ctd\u003e55\u003c/td\u003e\n      \u003ctd\u003e203\u003c/td\u003e\n      \u003ctd\u003e47.1\u003c/td\u003e\n      \u003ctd\u003e\u003ca href=\"https://drive.google.com/file/d/1yZHRDxa3eVvgDBBEwpCbsm_yqShEV3cC/view?usp=sharing\"\u003emodel\u003c/a\u003e \u003cbr/\u003e \u003ca href=\".assets/output_logs/r50_ms_smca_e50.txt\"\u003elog\u003c/a\u003e\u003c/td\u003e\n    \u003c/tr\u003e\n  \u003c/tbody\u003e\n\u003c/table\u003e\n\nNote:\n1. AP is computed on *COCO val2017*.\n2. \"DC5\" means removing the stride in C5 stage of ResNet and add a dilation of 2 instead.\n3. The GFLOPs of our models are estimated using [fvcore](https://github.com/facebookresearch/fvcore) on the first 100 images in *COCO val2017*. GFLOPs varies as input image sizes change. There may exist slight difference from actual values.\n\n\n## Citation\n\nIf you find SAM-DETR useful or inspiring, please consider citing:\n\n```bibtex\n@inproceedings{zhang2022-SAMDETR,\n  title      = {Accelerating {DETR} Convergence via Semantic-Aligned Matching},\n  author     = {Zhang, Gongjie and Luo, Zhipeng and Yu, Yingchen and Cui, Kaiwen and Lu, Shijian},\n  booktitle  = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},\n  pages      = {949-958},\n  year       = {2022},\n}\n```\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fmahshid1378%2Fmahshiddetr","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fmahshid1378%2Fmahshiddetr","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fmahshid1378%2Fmahshiddetr/lists"}