介绍
此榜单用于评比网上开源的目标检测模型;
模型入选标准
- 不使用额外的训练数据;
- 以开源代码的精度为准;
- 【数据源一】Paper with Code – SOTA
1 目标检测
COCO test-dev Benchmark Object Detection) | Papers With Code
Paper and Codes for COCO by 2022.1.20 & pause for update)
1. Dual-Swin-B-CBNetv2, boxAP: 60.1
模型: HTC-DB-Swin-L TTA)
1.1* Focal-L, boxAP: 58.9
Github-page: https://github.com/microsoft/Focal-Transformer
在其Github主页上未发现关于COCO数据集的精度数据,最高精度51.2;
(Focal-T-Cascade-Mask-R-CNN精度为51.5,不过会使用mask数据所以没有收录);
1.2* DyHead: 58.7
Github-page: DynamicHead
在其Github主页上最高精度为49.8,暂时不予收录;
2. Swin-L: 58.0val)
Github-page: Swin-L
在其Github主页上最高精度为58.0val);
3. YOLOR-D6*: 57.8
Github-page: YOLOR-D6*
4. SOLQ-{Swin-L & 1536}, boxAP: 56.5
模型:SOLQ-{Swin-L & 1536}
5. QueryInst , boxAP: 56.1
模型:QueryInst–Swin_L_300_queries–single_scale_testing
Note:
- 此榜单已经暂停更新,因为现在COCO-dev上SOTA的模型都是基于Transformer的,对于我们的项目来说没有实际意义,所以暂停此榜单的信息更新;
Real-Time Detection Models by 2021.12.28)
1. YOLOX-X, boxAP: 51.5, FPS: 57.8
模型:YOLOX-x
2. PP-YOLOv2, boxAP: 49.7, FPS: 49.5
模型:PP-YOLOv2–ResNet101vd
PP-YOLOv2是由Paddle推出的目标检测模型;
3. PicoDet-L, boxAP: 36.6, FPS: 45.8 21.85ms)
模型: PicoDet-L-416r
PicoDet是由Paddle推出的针对CPU的实时检测模型;
性能最好的是PicoDet-L-640r40.9-mAP), FPS为19.850.55ms)。
3. NanoDet-Plus-m-1.5x, boxAP: 34.1, FPS: 87.0 11.50ms)
模型: NanoDet-Plus-m-1.5x
Note:
- 这里的“Real-Time”指的是FPS在30以上的模型;
Look at Batch Size
Model | mAP | Bs single GPU) |
---|---|---|
YOLOX-s | 40.5 | 16 128/8 from paper) |
YOLOX-m | 47.2 | 16 |
YOLOX-l | 50.1 | 16 |
YOLOX-x | 51.5 | 16 |
YOLOX-Darknet53 | 48.0 | 16 |
PP-YOLOv2-ResNet50vd | 49.1 | 12 from code) |
PP-YOLOv2-ResNet101vd | 49.7 | 12 |
PicoDet-S-320r | 27.1 | 128 |
PicoDet-S-416r | 30.7 | 80 |
PicoDet-M-320r | 30.9 | 128 |
PicoDet-M-416r | 34.8 | 80 |
PicoDet-L-320r | 32.9 | 80 |
PicoDet-L-416r | 36.6 | 80 |
PicoDet-Shufflenetv2_1x | 30.0 | 80 |
PicoDet-MobileNetv3_large_1x | 35.6 | 80 |
PicoDet-LCNet-1.5x | 36.3 | 80 |
NanoDet-Plus-m-320r | 27.0 | 96 from code) |
NanoDet-Plus-m-416r | 30.4 | 96 |
NanoDet-Plus-m-1.5x-320r | 29.9 | 96 |
NanoDet-Plus-m-1.5x-416r | 34.1 | 96 |
2 语义分割
1. DeepLabV3Plus + SDCNetAug, MIoU: 83.5
3 图像分类
评测说明:
- 我们首先参考了 ImageNet Benchmark Image Classification) | Papers With Code
- ImageNet榜单除了参考paperswithcode之外,还参考了开源项目rwightman /
pytorch-image-models的实验结果 – Results – Pytorch Image Models
ImageNet Leaderboard by 2021.06.24)
1. SwinTransformer, top1: 87.148
Transformer-based
分类模型:swin_large_patch4_window12_384
2. CaiT-M-48-448, top1: 86.484
Transformer-based
分类模型:cait_m48_448
3. NFNet-F6 , top1: 86.296
分类模型:dm_nfnet_f6
CNN-based ImageNet Model by 2021.11.25)
1. MetaPseudoLabels EfficientNet-L2) , top1: 90.2%
分类模型:MetaPseudoLabels
*. VOLO: transformer-based
2. NFNet-F6+SAM, top1: 86.5%
分类模型:NFNet-F6+SAM
3. ResNeXt101 , top1: 86.26
分类模型:Fix_ResNeXt101_32x48d_wsl_paddle PaddleClas – EfficientNet and ResNeXt101_wsl series)
4. EfficientNetv2 , top1: 85.49
分类模型:tf_efficientnetv2_l