If you have any problems, suggestions or improvements, please submit the issue or PR.
[C^3 Framework] An open-source PyTorch code for crowd counting, which is released.
- Density Map Generation from Key Points [Matlab Code] [Python Code] [Fast Python Code]
- GCC Dataset [Link] (a large-scale, synthetic and diverse dataset)
- UCF-QNRF Dataset [Link / GoogleDrive]
- ShanghaiTech Dataset [Link: Dropbox / BaiduNetdisk]
- WorldExpo'10 Dataset [Link]
- UCF CC 50 Dataset [Link]
- Mall Dataset [Link]
- UCSD Dataset [Link]
- SmartCity Dataset [Link: GoogleDrive / BaiduNetdisk]
- AHU-Crowd Dataset [Link]
This section only includes the last ten papers since 2018 in arXiv.org. Previous papers will be hidden using <!--...-->. If you want to view them, please open the raw file to read the source code. Note that all unpublished arXiv papers are not included into the leaderboard of performance.
- Counting with Focus for Free [paper]
- Crowd Transformer Network [paper]
- W-Net: Reinforced U-Net for Density Map Estimation [paper]
- Crowd Counting with Decomposed Uncertainty [paper]
- Crowd Counting and Density Estimation by Trellis Encoder-Decoder Networks [paper]
- Generalizing semi-supervised generative adversarial networks to regression using feature contrasting [paper]
- Improving Dense Crowd Counting Convolutional Neural Networks using Inverse k-Nearest Neighbor Maps and Multiscale Upsampling [paper]
- Dual Path Multi-Scale Fusion Networks with Attention for Crowd Counting [paper]
- Scale-Aware Attention Network for Crowd Counting [paper]
- Mask-aware networks for crowd counting [paper]
- Context-Aware Crowd Counting [paper]
- PaDNet: Pan-Density Crowd Counting [paper]
- [CCWld] Learning from Synthetic Data for Crowd Counting in the Wild (CVPR2019) [paper] [Project] [arxiv])
- [SL2R] Exploiting Unlabeled Data in CNNs by Self-supervised Learning to Rank (T-PAMI) [paper](extension of L2R)
- [GWTA-CCNN] Almost Unsupervised Learning for Dense Crowd Counting (AAAI2019) [paper]
- [CAC] Class-Agnostic Counting (ACCV2018) [paper code]
- [L2R] Leveraging Unlabeled Data for Crowd Counting by Learning to Rank (CVPR2018) [paper] [code]
- [PACNN] Revisiting Perspective Information for Efficient Crowd Counting (CVPR2019)[paper]
- [PSDDN] Point in, Box out: Beyond Counting Persons in Crowds (CVPR2019)[paper]
- [ADCrowdNet] ADCrowdNet: An Attention-injective Deformable Convolutional Network for Crowd Understanding (CVPR2019) [paper]
- [CCWld] Learning from Synthetic Data for Crowd Counting in the Wild (CVPR2019) [paper] [Project] [arxiv]
- [SL2R] Exploiting Unlabeled Data in CNNs by Self-supervised Learning to Rank (T-PAMI) [paper](extension of L2R)
- [ASD] Adaptive Scenario Discovery for Crowd Counting (ICASSP2019) [paper]
- Crowd Counting Using Scale-Aware Attention Networks (WACV2019) [paper]
- [SPN] Scale Pyramid Network for Crowd Counting (WACV2019) [paper]
- [GWTA-CCNN] Almost Unsupervised Learning for Dense Crowd Counting (AAAI2019) [paper]
- Cross-Line Pedestrian Counting Based on Spatially-Consistent Two-Stage Local Crowd Density Estimation and Accumulation (CSVT2019) [paper]
- [LCFCN] Where are the Blobs: Counting by Localization with Point Supervision (ECCV2018) [paper] [code]
- [CAC] Class-Agnostic Counting (ACCV2018) [paper code]
- [SCNet] In Defense of Single-column Networks for Crowd Counting (BMVC2018) [paper]
- [AFP] Crowd Counting by Adaptively Fusing Predictions from an Image Pyramid (BMVC2018) [paper]
- [DRSAN] Crowd Counting using Deep Recurrent Spatial-Aware Network (IJCAI2018) [paper]
- [TDF-CNN] Top-Down Feedback for Crowd Counting Convolutional Neural Network (AAAI2018) [paper]
- [SANet] Scale Aggregation Network for Accurate and Efficient Crowd Counting (ECCV2018) [paper]
- [ic-CNN] Iterative Crowd Counting (ECCV2018) [paper]
- [CL] Composition Loss for Counting, Density Map Estimation and Localization in Dense Crowds (ECCV2018) [paper]
- [D-ConvNet] Crowd Counting with Deep Negative Correlation Learning (CVPR2018) [paper] [code]
- [IG-CNN] Divide and Grow: Capturing Huge Diversity in Crowd Images with Incrementally Growing CNN (CVPR2018) [paper]
- [BSAD] Body Structure Aware Deep Crowd Counting (TIP2018) [paper]
- [CSR] CSRNet: Dilated Convolutional Neural Networks for Understanding the Highly Congested Scenes (CVPR2018) [paper] [code]
- [L2R] Leveraging Unlabeled Data for Crowd Counting by Learning to Rank (CVPR2018) [paper] [code]
- [ACSCP] Crowd Counting via Adversarial Cross-Scale Consistency Pursuit (CVPR2018) [paper]
- [DecideNet] DecideNet: Counting Varying Density Crowds Through Attention Guided Detection and Density (CVPR2018) [paper]
- [AMDCN] An Aggregated Multicolumn Dilated Convolution Network for Perspective-Free Counting (CVPR2018) [paper] [code]
- [A-CCNN] A-CCNN: Adaptive CCNN for Density Estimation and Crowd Counting (ICIP2018) [paper]
- Crowd Counting with Fully Convolutional Neural Network (ICIP2018) [paper]
- [MS-GAN] Multi-scale Generative Adversarial Networks for Crowd Counting (ICPR2018) [paper]
- [DR-ResNet] A Deeply-Recursive Convolutional Network for Crowd Counting (ICASSP2018) [paper]
- [GAN-MTR] Crowd Counting With Minimal Data Using Generative Adversarial Networks For Multiple Target Regression (WACV2018) [paper]
- [SaCNN] Crowd counting via scale-adaptive convolutional neural network (WACV2018) [paper] [code]
- [Improved SaCNN] Improved Crowd Counting Method Based on Scale-Adaptive Convolutional Neural Network (IEEE Access) [paper]
- [DA-Net] DA-Net: Learning the Fine-Grained Density Distribution With Deformation Aggregation Network (IEEE Access) [paper][code]
- [NetVLAD] Multiscale Multitask Deep NetVLAD for Crowd Counting (TII2018) [paper] [code]
- [W-VLAD] Crowd Counting via Weighted VLAD on Dense Attribute Feature Maps (CSVT2018) [paper]
- [CP-CNN] Generating High-Quality Crowd Density Maps using Contextual Pyramid CNNs (ICCV2017) [paper]
- [ConvLSTM] Spatiotemporal Modeling for Crowd Counting in Videos (ICCV2017) [paper]
- [CMTL] CNN-based Cascaded Multi-task Learning of High-level Prior and Density Estimation for Crowd Counting (AVSS2017) [paper] [code]
- [ResnetCrowd] ResnetCrowd: A Residual Deep Learning Architecture for Crowd Counting, Violent Behaviour Detection and Crowd Density Level Classification (AVSS2017) [paper]
- [Switching CNN] Switching Convolutional Neural Network for Crowd Counting (CVPR2017) [paper] [code]
- A Survey of Recent Advances in CNN-based Single Image Crowd Counting and Density Estimation (PR Letters2017) [paper]
- [DAL-SVR] Boosting deep attribute learning via support vector regression for fast moving crowd counting (PR Letters2017) [paper]
- [MSCNN] Multi-scale Convolution Neural Networks for Crowd Counting (ICIP2017) [paper] [code]
- [FCNCC] Fully Convolutional Crowd Counting On Highly Congested Scenes (VISAPP2017) [paper]
- [Hydra-CNN] Towards perspective-free object counting with deep learning (ECCV2016) [paper] [code]
- [CNN-Boosting] Learning to Count with CNN Boosting (ECCV2016) [paper]
- [Crossing-line] Crossing-line Crowd Counting with Two-phase Deep Neural Networks (ECCV2016) [paper]
- [CrowdNet] CrowdNet: A Deep Convolutional Network for Dense Crowd Counting (ACMMM2016) [paper] [code]
- [MCNN] Single-Image Crowd Counting via Multi-Column Convolutional Neural Network (CVPR2016) [paper] [unofficial code: TensorFlow PyTorch]
- [Shang 2016] End-to-end crowd counting via joint learning local and global count (ICIP2016) [paper]
- [RPF] Crowd Density Estimation based on Rich Features and Random Projection Forest (WACV2016) [paper]
- [CS-SLR] Cost-sensitive sparse linear regression for crowd counting with imbalanced training data (ICME2016) [paper]
- [Faster-OHEM-KCF] Deep People Counting with Faster R-CNN and Correlation Tracking (ICME2016) [paper]
- [COUNT Forest] COUNT Forest: CO-voting Uncertain Number of Targets using Random Forest for Crowd Density Estimation (ICCV2015) [paper]
- [Bayesian] Bayesian Model Adaptation for Crowd Counts (ICCV2015) [paper]
- [Zhang 2015] Cross-scene Crowd Counting via Deep Convolutional Neural Networks (CVPR2015) [paper] [code]
- [Wang 2015] Deep People Counting in Extremely Dense Crowds (ACMMM2015) [paper]
- [Fu 2015] Fast crowd density estimation with convolutional neural networks (AI2015) [paper]
- [Idrees 2013] Multi-Source Multi-Scale Counting in Extremely Dense Crowd Images (CVPR2013) [paper]
- [Ma 2013] Crossing the Line: Crowd Counting by Integer Programming with Local Features (CVPR2013) [paper]
- [Chen 2013] Feature mining for localised crowd counting (BMVC2012) [paper]
- [Lempitsky 2010] Learning To Count Objects in Images (NIPS2010) [paper]
- [Chan 2008] Privacy preserving crowd monitoring: Counting people without people models or tracking (CVPR 2008) [paper]
The section is being continually updated. Note that some values have superscript, which indicates their source.
| Year-Conference/Journal | Methods | MAE | MSE | PSNR | SSIM | Params | Pre-trained Model |
|---|---|---|---|---|---|---|---|
| 2016--CVPR | MCNN | 110.2 | 173.2 | 21.4CSR | 0.52CSR | 0.13MSANet | None |
| 2017--ICIP | MSCNN | 83.8 | 127.4 | - | - | - | - |
| 2017--AVSS | CMTL | 101.3 | 152.4 | - | - | - | None |
| 2017--CVPR | Switching CNN | 90.4 | 135.0 | - | - | - | VGG-16 |
| 2017--ICCV | CP-CNN | 73.6 | 106.4 | - | - | - | - |
| 2018--WACV | SaCNN | 86.8 | 139.2 | - | - | - | - |
| 2018--CVPR | ACSCP | 75.7 | 102.7 | - | - | 5.1M | None |
| 2018--CVPR | CSRNet | 68.2 | 115.0 | 23.79 | 0.76 | 16.26MSANet | VGG-16 |
| 2018--CVPR | IG-CNN | 72.5 | 118.2 | - | - | - | - |
| 2018--CVPR | D-ConvNet-v1 | 73.5 | 112.3 | - | - | - | - |
| 2018--CVPR | L2R (Multi-task, Query-by-example) | 72.0 | 106.6 | - | - | - | VGG-16 |
| 2018--CVPR | L2R (Multi-task, Keyword) | 73.6 | 112.0 | - | - | - | VGG-16 |
| 2018--IJCAI | DRSAN | 69.3 | 96.4 | - | - | - | - |
| 2018--ECCV | ic-CNN (one stage) | 69.8 | 117.3 | - | - | - | - |
| 2018--ECCV | ic-CNN (two stages) | 68.5 | 116.2 | - | - | - | - |
| 2018--ECCV | SANet | 67.0 | 104.5 | - | - | 0.91M | None |
| 2018--AAAI | TDF-CNN | 97.5 | 145.1 | - | - | - | - |
| 2019--AAAI | GWTA-CCNN | 154.7 | 229.4 | - | - | - | - |
| 2019--ICASSP | ASD | 65.6 | 98.0 | - | - | - | - |
| 2019--CVPR | SFCN | 64.8 | 107.5 | - | - | - | - |
| Year-Conference/Journal | Methods | MAE | MSE |
|---|---|---|---|
| 2016--CVPR | MCNN | 26.4 | 41.3 |
| 2017--ICIP | MSCNN | 17.7 | 30.2 |
| 2017--AVSS | CMTL | 20.0 | 31.1 |
| 2017--CVPR | Switching CNN | 21.6 | 33.4 |
| 2017--ICCV | CP-CNN | 20.1 | 30.1 |
| 2018--TIP | BSAD | 20.2 | 35.6 |
| 2018--WACV | SaCNN | 16.2 | 25.8 |
| 2018--CVPR | ACSCP | 17.2 | 27.4 |
| 2018--CVPR | CSRNet | 10.6 | 16.0 |
| 2018--CVPR | IG-CNN | 13.6 | 21.1 |
| 2018--CVPR | D-ConvNet-v1 | 18.7 | 26.0 |
| 2018--CVPR | DecideNet | 21.53 | 31.98 |
| 2018--CVPR | DecideNet + R3 | 20.75 | 29.42 |
| 2018--CVPR | L2R (Multi-task, Query-by-example) | 14.4 | 23.8 |
| 2018--CVPR | L2R (Multi-task, Keyword) | 13.7 | 21.4 |
| 2018--IJCAI | DRSAN | 11.1 | 18.2 |
| 2018--AAAI | TDF-CNN | 20.7 | 32.8 |
| 2018--ECCV | ic-CNN (one stage) | 10.4 | 16.7 |
| 2018--ECCV | ic-CNN (two stages) | 10.7 | 16.0 |
| 2018--ECCV | SANet | 8.4 | 13.6 |
| 2019--ICASSP | ASD | 8.5 | 13.7 |
| 2019--CVPR | SFCN | 7.6 | 13.0 |
| Year-Conference/Journal | Method | C-MAE | C-NAE | C-MSE | DM-MAE | DM-MSE | DM-HI | L- Av. Precision | L-Av. Recall | L-AUC |
|---|---|---|---|---|---|---|---|---|---|---|
| 2013--CVPR | Idrees 2013CL | 315 | 0.63 | 508 | - | - | - | - | - | - |
| 2016--CVPR | MCNNCL | 277 | 0.55 | 0.006670 | 0.0223 | 0.5354 | 59.93% | 63.50% | 0.591 | |
| 2017--AVSS | CMTLCL | 252 | 0.54 | 514 | 0.005932 | 0.0244 | 0.5024 | - | - | - |
| 2017--CVPR | Switching CNNCL | 228 | 0.44 | 445 | 0.005673 | 0.0263 | 0.5301 | - | - | - |
| 2018--ECCV | CL | 132 | 0.26 | 191 | 0.00044 | 0.0017 | 0.9131 | 75.8% | 59.75% | 0.714 |
| 2019--CVPR | SFCN | 102.0 | - | 171.4 | - | - | - | - | - | - |
| Year-Conference/Journal | Methods | MAE | MSE |
|---|---|---|---|
| 2013--CVPR | Idrees 2013 | 468.0 | 590.3 |
| 2015--CVPR | Zhang 2015 | 467.0 | 498.5 |
| 2016--CVPR | MCNN | 377.6 | 509.1 |
| 2016--ACM MM | CrowdNet | 452.5 | - |
| 2016--ECCV | Hydra-CNN | 333.73 | 425.26 |
| 2016--ECCV | CNN-Boosting | 364.4 | - |
| 2016--ICIP | Shang 2016 | 270.3 | - |
| 2017--ICIP | MSCNN | 363.7 | 468.4 |
| 2017--AVSS | CMTL | 322.8 | 397.9 |
| 2017--CVPR | Switching CNN | 318.1 | 439.2 |
| 2017--ICCV | ConvLSTM-nt | 284.5 | 297.1 |
| 2017--ICCV | CP-CNN | 298.8 | 320.9 |
| 2018--TIP | BSAD | 409.5 | 563.7 |
| 2018--WACV | SaCNN | 314.9 | 424.8 |
| 2018--CVPR | ACSCP | 291.0 | 404.6 |
| 2018--CVPR | CSRNet | 266.1 | 397.5 |
| 2018--CVPR | IG-CNN | 291.4 | 349.4 |
| 2018--CVPR | D-ConvNet-v1 | 288.4 | 404.7 |
| 2018--CVPR | L2R (Multi-task, Query-by-example) | 291.5 | 397.6 |
| 2018--CVPR | L2R (Multi-task, Keyword) | 279.6 | 388.9 |
| 2018--IJCAI | DRSAN | 219.2 | 250.2 |
| 2018--ECCV | ic-CNN (two stages) | 260.9 | 365.5 |
| 2018--ECCV | SANet | 258.4 | 334.9 |
| 2018--AAAI | TDF-CNN | 354.7 | 491.4 |
| 2019--AAAI | GWTA-CCNN | 433.7 | 583.3 |
| 2019--CVPR | SFCN | 214.2 | 318.2 |
| 2019--ICASSP | ASD | 196.2 | 270.9 |
| Year-Conference/Journal | Method | S1 | S2 | S3 | S4 | S5 | Avg. |
|---|---|---|---|---|---|---|---|
| 2015--CVPR | Zhang 2015 | 9.8 | 14.1 | 14.3 | 22.2 | 3.7 | 12.9 |
| 2016--CVPR | MCNN | 3.4 | 20.6 | 12.9 | 13.0 | 8.1 | 11.6 |
| 2017--ICIP | MSCNN | 7.8 | 15.4 | 14.9 | 11.8 | 5.8 | 11.7 |
| 2017--CVPR | Switching CNN | 4.4 | 15.7 | 10.0 | 11.0 | 5.9 | 9.4 |
| 2017--ICCV | ConvLSTM-nt | 8.6 | 16.9 | 14.6 | 15.4 | 4.0 | 11.9 |
| 2017--ICCV | ConvLSTM | 7.1 | 15.2 | 15.2 | 13.9 | 3.5 | 10.9 |
| 2017--ICCV | Bidirectional ConvLSTM | 6.8 | 14.5 | 14.9 | 13.5 | 3.1 | 10.6 |
| 2017--ICCV | CP-CNN | 2.9 | 14.7 | 10.5 | 10.4 | 5.8 | 8.86 |
| 2018--TIP | BSAD | 4.1 | 21.7 | 11.9 | 11.0 | 3.5 | 10.5 |
| 2018--WACV | SaCNN | 2.6 | 13.5 | 10.6 | 12.5 | 3.3 | 8.5 |
| 2018--CVPR | ACSCP | 2.8 | 14.05 | 9.6 | 8.1 | 2.9 | 7.5 |
| 2018--CVPR | CSRNet | 2.9 | 11.5 | 8.6 | 16.6 | 3.4 | 8.6 |
| 2018--CVPR | IG-CNN | 2.6 | 16.1 | 10.15 | 20.2 | 7.6 | 11.3 |
| 2018--CVPR | D-ConvNet-v1 | 1.9 | 12.1 | 20.7 | 8.3 | 2.6 | 9.1 |
| 2018--CVPR | DecideNet | 2.0 | 13.14 | 8.9 | 17.4 | 4.75 | 9.23 |
| 2018--IJCAI | DRSAN | 2.6 | 11.8 | 10.3 | 10.4 | 3.7 | 7.76 |
| 2018--AAAI | TDF-CNN | 2.7 | 23.4 | 10.7 | 17.6 | 3.3 | 11.5 |
| 2018--ECCV | ic-CNN | 17.0 | 12.3 | 9.2 | 8.1 | 4.7 | 10.3 |
| 2018--ECCV | SANet | 2.6 | 13.2 | 9.0 | 13.3 | 3.0 | 8.2 |
| 2019--CVPR | PACNN | 2.3 | 12.5 | 9.1 | 11.2 | 3.8 | 7.8 |
| 2019--CVPR | ADCrowdNet(AMG-bAttn-DME) | 1.7 | 14.4 | 11.5 | 7.9 | 3.0 | 7.7 |
| 2019--CVPR | ADCrowdNet(AMG-attn-DME) | 1.6 | 13.2 | 8.7 | 10.6 | 2.6 | 7.3 |
| Year-Conference/Journal | Method | MAE | MSE |
|---|---|---|---|
| 2015--CVPR | Zhang 2015 | 1.60 | 3.31 |
| 2016--CVPR | MCNN | 1.07 | 1.35 |
| 2016--ECCV | Hydra-CNN | 1.65 | - |
| 2016--ECCV | CNN-Boosting | 1.10 | - |
| 2017--CVPR | Switching CNN | 1.62 | 2.10 |
| 2017--ICCV | ConvLSTM-nt | 1.73 | 3.52 |
| 2017--ICCV | ConvLSTM | 1.30 | 1.79 |
| 2017--ICCV | Bidirectional ConvLSTM | 1.13 | 1.43 |
| 2018--TIP | BSAD | 1.00 | 1.40 |
| 2018--CVPR | ACSCP | 1.04 | 1.35 |
| 2018--CVPR | CSRNet | 1.16 | 1.47 |
| 2018--ECCV | SANet | 1.02 | 1.29 |
| 2019--CVPR | ADCrowdNet | 0.98 | 1.25 |
| 2019--CVPR | PACNN | 0.89 | 1.18 |