PyTorch Person Re-identification Models

We collected some ReID models from open-source projects and provided a unified interface to access them. We retrained these models on three datasets: Market1501, DukeMTMCReID, and MSMT17 using the original open-source code and can automatically load these trained model weights at runtime. For specific usage examples, see test.py.

The main directory structure

.
├── reid_models
│   ├── data # data load module
│   │   ├── build.py # dataset and dataloader building function
│   │   ├── datasets
│   │   ├── samplers
│   │   └── transforms
│   ├── evaluate # metrics evaluation module
│   │   ├── estimator.py
│   │   ├── eval_function.py # cuda-accelerated reid evaluation function
│   │   └── matcher.py
│   └── modeling # model building module
│       ├── build.py # model building and loading trained weights function
│       ├── models_config.yaml # model configs file
│       └── third_party_models # third-party open source reid models
│           ├── ABDNet
│           ├── AGWNet
│           ├── APNet
│           ├── DeepPersonReid
│           ├── FastReID
│           ├── ReidStrongBaseline
│           └── TransReID
├── tools # some useful tools
│   ├── imagenet_pretrain.py # imagent data pre-training
│   ├── preprocessing_imagenet.py # offline resize image, for faster data loading
│   ├── visualizer_ranklist.py # visualize match results
│   ├── visualizer_ranklist_with_gradcam.py # # visualize match results with gradcam
│   └── visualizer_tsne.py # visualize tsne
├── train.py # reid model training
├── test.py # reid model testing

Highlights

We rewrite a CUDA-accelerated version of the ReID evaluation code, which has a huge speedup compared to the previous numpy version. For details, see reid_models/evaluate/eval_function.py.
For each third-party ReID project, we provide a way to dynamically register the model, as seen reid_models/modeling/third_party_models/FastReID/__init__.py. We will expand more models in the future.
We provide some useful visualization, training and testing tools, as seen in tools
We provide an efficient pre-training code for the ImageNet1k dataset, based on the accelerate library, as see in tools/imagenet_pretrain.py

Configuration files

For path settings with datasets, see reid_models/data/datasets/dataset_paths.yaml
For model parameters setting, see reid_models/modeling/models_config.yaml
For fast-reid models settings, see reid_models/modeling/third_party_models/FastReID/configs

Visualization

Visual matching samples

Visual matching samples with gradcam

Visual TSNE

Model metrics

Dataset	Model	Top1	Top5	mAP	mINP	MACs	Params
dukemtmcreid	densenet121_abd	0.885	0.940	0.775	0.410	5.619G	37.569M
market1501	densenet121_abd	0.953	0.984	0.876	0.639	5.619G	37.719M
msmt17	densenet121_abd	0.817	0.903	0.591	0.151	5.620G	38.611M
dukemtmcreid	resnet50_abd	0.878	0.941	0.773	0.398	9.377G	69.025M
market1501	resnet50_abd	0.950	0.983	0.872	0.634	9.377G	69.175M
msmt17	resnet50_abd	0.800	0.897	0.579	0.143	9.378G	70.067M
dukemtmcreid	resnet50_agw	0.892	0.951	0.796	0.454	4.094G	23.541M
market1501	resnet50_agw	0.955	0.983	0.883	0.653	4.094G	23.541M
msmt17	resnet50_agw	0.789	0.885	0.554	0.128	4.094G	23.541M
dukemtmcreid	resnet50_ap	0.897	0.954	0.800	0.457	16.357G	51.281M
market1501	resnet50_ap	0.952	0.985	0.890	0.676	16.357G	51.281M
msmt17	resnet50_ap	0.792	0.892	0.575	0.145	16.357G	51.281M
dukemtmcreid	mlfn_dpr	0.809	0.903	0.642	0.250	2.812G	32.473M
market1501	mlfn_dpr	0.897	0.960	0.747	0.423	2.812G	32.473M
msmt17	mlfn_dpr	0.673	0.803	0.390	0.059	2.812G	32.473M
dukemtmcreid	osnet_x1_0_dpr	0.878	0.938	0.764	0.403	1.013G	2.170M
market1501	osnet_x1_0_dpr	0.950	0.980	0.865	0.614	1.013G	2.170M
msmt17	osnet_x1_0_dpr	0.784	0.882	0.547	0.128	1.013G	2.170M
dukemtmcreid	osnet_ain_x1_0_dpr	0.869	0.943	0.744	0.364	1.013G	2.170M
market1501	osnet_ain_x1_0_dpr	0.933	0.975	0.836	0.551	1.013G	2.170M
msmt17	osnet_ain_x1_0_dpr	0.765	0.871	0.508	0.101	1.013G	2.170M
dukemtmcreid	osnet_ibn_x1_0_dpr	0.869	0.941	0.743	0.361	1.017G	2.171M
market1501	osnet_ibn_x1_0_dpr	0.939	0.976	0.836	0.553	1.017G	2.171M
msmt17	osnet_ibn_x1_0_dpr	0.767	0.871	0.503	0.095	1.017G	2.171M
dukemtmcreid	resnet50_bot	0.860	0.938	0.765	0.402	4.087G	23.512M
market1501	resnet50_bot	0.941	0.982	0.857	0.593	4.087G	23.512M
msmt17	resnet50_bot	0.739	0.854	0.503	0.106	4.087G	23.512M
dukemtmcreid	resnet50_ibn_a_bot	0.889	0.952	0.792	0.445	4.087G	23.512M
market1501	resnet50_ibn_a_bot	0.952	0.986	0.872	0.633	4.087G	23.512M
msmt17	resnet50_ibn_a_bot	0.792	0.885	0.565	0.135	4.087G	23.512M
dukemtmcreid	se_resnet50_bot	0.868	0.943	0.768	0.407	3.992G	26.043M
market1501	se_resnet50_bot	0.944	0.982	0.864	0.607	3.992G	26.043M
msmt17	se_resnet50_bot	0.687	0.819	0.460	0.092	3.992G	26.043M
dukemtmcreid	se_resnext50_bot	0.884	0.950	0.786	0.433	4.104G	25.515M
market1501	se_resnext50_bot	0.949	0.985	0.879	0.645	4.104G	25.515M
msmt17	se_resnext50_bot	0.768	0.869	0.546	0.129	4.104G	25.515M
dukemtmcreid	senet154_bot	0.880	0.950	0.782	0.428	17.136G	113.044M
market1501	senet154_bot	0.946	0.981	0.865	0.612	17.136G	113.044M
msmt17	senet154_bot	0.812	0.900	0.584	0.142	17.136G	113.044M
dukemtmcreid	deit_transreid	0.907	0.958	0.819	0.486	-	-
market1501	deit_transreid	0.950	0.983	0.885	0.672	-	-
msmt17	deit_transreid	0.840	0.919	0.663	0.238	-	-
dukemtmcreid	vit_base_transreid	0.890	0.954	0.796	0.454	11.042G	85.648M
market1501	vit_base_transreid	0.946	0.982	0.871	0.641	11.042G	85.648M
msmt17	vit_base_transreid	0.817	0.906	0.618	0.204	11.042G	85.648M
dukemtmcreid	vit_transreid	0.908	0.960	0.821	0.503	-	-
market1501	vit_transreid	0.951	0.984	0.890	0.693	-	-
msmt17	vit_transreid	0.853	0.925	0.678	0.257	-	-
dukemtmcreid	agw_R50_fastreid	0.891	0.948	0.791	0.445	4.057G	23.475M
market1501	agw_R50_fastreid	0.953	0.987	0.883	0.654	4.057G	23.475M
msmt17	agw_R50_fastreid	0.792	0.883	0.557	0.129	4.057G	23.475M
dukemtmcreid	agw_R50_ibn_fastreid	0.908	0.955	0.806	0.471	4.059G	23.478M
market1501	agw_R50_ibn_fastreid	0.955	0.982	0.889	0.674	4.059G	23.478M
msmt17	agw_R50_ibn_fastreid	0.818	0.903	0.595	0.155	4.059G	23.478M
dukemtmcreid	agw_R101_ibn_fastreid	0.908	0.953	0.810	0.476	6.488G	42.453M
market1501	agw_R101_ibn_fastreid	0.953	0.985	0.891	0.682	6.488G	42.453M
msmt17	agw_R101_ibn_fastreid	0.825	0.906	0.611	0.167	6.488G	42.453M
dukemtmcreid	agw_S50_fastreid	0.907	0.957	0.811	0.475	4.665G	25.370M
market1501	agw_S50_fastreid	0.952	0.983	0.892	0.684	4.665G	25.370M
msmt17	agw_S50_fastreid	0.838	0.915	0.646	0.192	4.665G	25.370M
dukemtmcreid	bagtricks_mobilenet_v3_large_fastreid	0.823	0.920	0.691	0.299	179.652M	4.176M
market1501	bagtricks_mobilenet_v3_large_fastreid	0.914	0.973	0.790	0.450	179.652M	4.176M
msmt17	bagtricks_mobilenet_v3_large_fastreid	0.680	0.814	0.426	0.069	179.652M	4.176M
dukemtmcreid	bagtricks_osnet_ibn_x1_0_fastreid	0.876	0.943	0.750	0.384	986.660M	1.885M
market1501	bagtricks_osnet_ibn_x1_0_fastreid	0.936	0.977	0.841	0.561	986.660M	1.885M
msmt17	bagtricks_osnet_ibn_x1_0_fastreid	0.782	0.884	0.527	0.108	986.660M	1.885M
dukemtmcreid	bagtricks_osnet_x1_0_fastreid	0.879	0.945	0.769	0.405	980.369M	1.884M
market1501	bagtricks_osnet_x1_0_fastreid	0.941	0.979	0.863	0.610	980.369M	1.884M
msmt17	bagtricks_osnet_x1_0_fastreid	0.802	0.896	0.570	0.138	980.369M	1.884M
dukemtmcreid	bagtricks_R50_fastreid	0.869	0.946	0.766	0.402	4.053G	23.455M
market1501	bagtricks_R50_fastreid	0.945	0.982	0.861	0.600	4.053G	23.455M
msmt17	bagtricks_R50_fastreid	0.752	0.862	0.516	0.111	4.053G	23.455M
dukemtmcreid	bagtricks_SeR50_fastreid	0.858	0.937	0.751	0.386	4.060G	25.970M
market1501	bagtricks_SeR50_fastreid	0.942	0.982	0.856	0.594	4.060G	25.970M
msmt17	bagtricks_SeR50_fastreid	0.742	0.855	0.505	0.107	4.060G	25.970M
dukemtmcreid	bagtricks_R50_ibn_fastreid	0.899	0.952	0.788	0.437	4.056G	23.457M
market1501	bagtricks_R50_ibn_fastreid	0.954	0.982	0.879	0.647	4.056G	23.457M
msmt17	bagtricks_R50_ibn_fastreid	0.794	0.889	0.565	0.137	4.056G	23.457M
dukemtmcreid	bagtricks_R101_ibn_fastreid	0.897	0.953	0.796	0.451	6.481G	42.401M
market1501	bagtricks_R101_ibn_fastreid	0.955	0.983	0.886	0.674	6.481G	42.401M
msmt17	bagtricks_R101_ibn_fastreid	0.815	0.900	0.593	0.153	6.481G	42.401M
dukemtmcreid	bagtricks_S50_fastreid	0.897	0.953	0.797	0.455	4.665G	25.370M
market1501	bagtricks_S50_fastreid	0.952	0.983	0.886	0.668	4.665G	25.370M
msmt17	bagtricks_S50_fastreid	0.818	0.905	0.607	0.169	4.665G	25.370M
dukemtmcreid	bagtricks_convnext_tiny_fastreid	0.807	0.898	0.657	0.260	4.248G	26.630M
market1501	bagtricks_convnext_tiny_fastreid	0.906	0.967	0.768	0.431	4.248G	26.630M
msmt17	bagtricks_convnext_tiny_fastreid	0.644	0.784	0.380	0.057	4.248G	26.630M
dukemtmcreid	bagtricks_densenet121_fastreid	0.853	0.935	0.727	0.335	1.850G	6.870M
market1501	bagtricks_densenet121_fastreid	0.926	0.975	0.820	0.511	1.850G	6.870M
msmt17	bagtricks_densenet121_fastreid	0.729	0.850	0.473	0.084	1.850G	6.870M
dukemtmcreid	bagtricks_inception_resnet_v2_fastreid	0.846	0.924	0.699	0.307	3.784G	54.306M
market1501	bagtricks_inception_resnet_v2_fastreid	0.922	0.973	0.789	0.427	3.784G	54.306M
msmt17	bagtricks_inception_resnet_v2_fastreid	0.703	0.829	0.438	0.072	3.784G	54.306M
dukemtmcreid	bagtricks_inception_v3_fastreid	0.829	0.914	0.677	0.284	1.681G	21.786M
market1501	bagtricks_inception_v3_fastreid	0.918	0.971	0.789	0.462	1.681G	21.786M
msmt17	bagtricks_inception_v3_fastreid	0.666	0.794	0.388	0.056	1.681G	21.786M
dukemtmcreid	bagtricks_inception_v4_fastreid	0.809	0.896	0.658	0.237	3.630G	41.143M
market1501	bagtricks_inception_v4_fastreid	0.894	0.957	0.749	0.402	3.630G	41.143M
msmt17	bagtricks_inception_v4_fastreid	0.614	0.766	0.342	0.039	3.630G	41.143M
dukemtmcreid	sbs_R50_fastreid	0.904	0.952	0.798	0.453	4.057G	23.475M
market1501	sbs_R50_fastreid	0.957	0.984	0.881	0.653	4.057G	23.475M
msmt17	sbs_R50_fastreid	0.822	0.906	0.583	0.134	4.057G	23.475M
dukemtmcreid	sbs_R50_ibn_fastreid	0.902	0.952	0.800	0.449	4.059G	23.478M
market1501	sbs_R50_ibn_fastreid	0.954	0.984	0.883	0.657	4.059G	23.478M
msmt17	sbs_R50_ibn_fastreid	0.830	0.908	0.592	0.140	4.059G	23.478M
dukemtmcreid	sbs_R101_ibn_fastreid	0.911	0.955	0.815	0.468	6.488G	42.453M
market1501	sbs_R101_ibn_fastreid	0.956	0.985	0.894	0.687	6.488G	42.453M
msmt17	sbs_R101_ibn_fastreid	0.825	0.902	0.581	0.131	6.488G	42.453M
dukemtmcreid	sbs_S50_fastreid	0.908	0.955	0.798	0.429	4.665G	25.370M
market1501	sbs_S50_fastreid	0.957	0.985	0.883	0.657	4.665G	25.370M
msmt17	sbs_S50_fastreid	0.834	0.911	0.603	0.151	4.665G	25.370M
dukemtmcreid	mgn_R50_fastreid	0.882	0.941	0.786	0.426	9.309G	68.675M
market1501	mgn_R50_fastreid	0.944	0.981	0.872	0.636	9.309G	68.675M
msmt17	mgn_R50_fastreid	0.786	0.883	0.569	0.142	9.309G	68.675M
dukemtmcreid	mgn_R50_ibn_fastreid	0.887	0.947	0.785	0.432	9.312G	68.680M
market1501	mgn_R50_ibn_fastreid	0.942	0.979	0.876	0.649	9.312G	68.680M
msmt17	mgn_R50_ibn_fastreid	0.792	0.883	0.585	0.167	9.312G	68.680M
dukemtmcreid	mgn_sbs_R50_fastreid	0.892	0.950	0.802	0.450	9.309G	68.675M
market1501	mgn_sbs_R50_fastreid	0.955	0.989	0.886	0.644	9.309G	68.675M
msmt17	mgn_sbs_R50_fastreid	0.830	0.914	0.616	0.152	9.309G	68.675M
dukemtmcreid	mgn_sbs_R50_ibn_fastreid	0.909	0.960	0.814	0.460	9.312G	68.680M
market1501	mgn_sbs_R50_ibn_fastreid	0.958	0.986	0.893	0.667	9.312G	68.680M
msmt17	mgn_sbs_R50_ibn_fastreid	0.853	0.924	0.652	0.184	9.312G	68.680M
dukemtmcreid	mgn_agw_R50_fastreid	0.887	0.942	0.792	0.441	9.309G	68.675M
market1501	mgn_agw_R50_fastreid	0.948	0.981	0.877	0.646	9.309G	68.675M
msmt17	mgn_agw_R50_fastreid	0.798	0.891	0.590	0.156	9.309G	68.675M
dukemtmcreid	mgn_agw_R50_ibn_fastreid	0.898	0.947	0.789	0.442	9.312G	68.680M
market1501	mgn_agw_R50_ibn_fastreid	0.939	0.977	0.870	0.635	9.312G	68.680M
msmt17	mgn_agw_R50_ibn_fastreid	0.797	0.885	0.593	0.173	9.312G	68.680M
dukemtmcreid	mgn_S50_fastreid	0.873	0.938	0.763	0.406	9.924G	73.946M
market1501	mgn_S50_fastreid	0.942	0.978	0.861	0.620	9.924G	73.946M
msmt17	mgn_S50_fastreid	0.782	0.879	0.566	0.165	9.924G	73.946M
dukemtmcreid	mgn_S50_ibn_fastreid	0.880	0.938	0.761	0.400	9.924G	73.946M
market1501	mgn_S50_ibn_fastreid	0.941	0.976	0.863	0.625	9.924G	73.946M
msmt17	mgn_S50_ibn_fastreid	0.777	0.877	0.563	0.161	9.924G	73.946M
dukemtmcreid	mgn_sbs_S50_fastreid	0.921	0.958	0.822	0.469	9.924G	73.946M
market1501	mgn_sbs_S50_fastreid	0.954	0.985	0.895	0.680	9.924G	73.946M
msmt17	mgn_sbs_S50_fastreid	0.863	0.928	0.664	0.190	9.924G	73.946M
dukemtmcreid	mgn_sbs_S50_ibn_fastreid	0.915	0.957	0.821	0.472	9.924G	73.946M
market1501	mgn_sbs_S50_ibn_fastreid	0.959	0.986	0.895	0.675	9.924G	73.946M
msmt17	mgn_sbs_S50_ibn_fastreid	0.863	0.924	0.658	0.183	9.924G	73.946M
dukemtmcreid	mgn_agw_S50_fastreid	0.888	0.943	0.782	0.426	9.924G	73.946M
market1501	mgn_agw_S50_fastreid	0.944	0.975	0.869	0.639	9.924G	73.946M
msmt17	mgn_agw_S50_fastreid	0.798	0.886	0.601	0.182	9.924G	73.946M
dukemtmcreid	mgn_agw_S50_ibn_fastreid	0.892	0.942	0.787	0.429	9.924G	73.946M
market1501	mgn_agw_S50_ibn_fastreid	0.946	0.979	0.867	0.633	9.924G	73.946M
msmt17	mgn_agw_S50_ibn_fastreid	0.799	0.887	0.603	0.180	9.924G	73.946

Acknowledge

[light-reid] https://github.com/wangguanan/light-reid
[ABD-Net] https://github.com/VITA-Group/ABD-Net
[AGW-Net] https://github.com/mangye16/ReID-Survey/
[AP-Net] https://github.com/CHENGY12/APNet
[DeepPersonReid] https://github.com/KaiyangZhou/deep-person-reid
[Fast-ReID] https://github.com/JDAI-CV/fast-reid
[ReidStrongBaseline] https://github.com/michuanhaohao/reid-strong-baseline
[TransReID] https://github.com/damo-cv/TransReID

HWliiu / pytorch-reid-models