1. 量化#
1.1 图象分类#
数据集：ImageNet1000类
| Model | Top-1/Top-5 | 模型大小（MB） | 下载 |
|---|---|---|---|
| MobileNetV1 FP32 | 70.99%/89.68% | xx | 下载链接 |
| MobileNetV1 quant_post | xx%/xx% | xx | 下载链接 |
| MobileNetV1 quant_aware | xx%/xx% | xx | 下载链接 |
| MobileNetV2 FP32 | 72.15%/90.65% | xx | 下载链接 |
| MobileNetV2 quant_post | xx%/xx% | xx | 下载链接 |
| MobileNetV2 quant_aware | xx%/xx% | xx | 下载链接 |
| ResNet50 FP32 | 76.50%/93.00% | xx | 下载链接 |
| ResNet50 quant_post | xx%/xx% | xx | 下载链接 |
| ResNet50 quant_aware | xx%/xx% | xx | 下载链接 |
1.2 目标检测#
数据集：COCO 2017
| Model | Image/GPU | 输入608 Box AP | 输入416 Box AP | 输入320 Box AP | 模型大小(MB) | 下载 |
|---|---|---|---|---|---|---|
| MobileNet-V1-YOLOv3 FP32 | 8 | 29.3 | 29.3 | 27.1 | xx | 下载链接 |
| MobileNet-V1-YOLOv3 quant_post | 8 | xx | xx | xx | xx | 下载链接 |
| MobileNet-V1-YOLOv3 quant_aware | 8 | xx | xx | xx | xx | 下载链接 |
| R50-dcn-YOLOv3 FP32 | 8 | 41.4 | - | - | xx | 下载链接 |
| R50-dcn-YOLOv3 quant_post | 8 | xx | - | - | xx | 下载链接 |
| R50-dcn-YOLOv3 quant_aware | 8 | xx | - | - | xx | 下载链接 |
数据集：WIDER-FACE
| Model | Image/GPU | 输入尺寸 | Easy/Medium/Hard | 模型大小（MB） | 下载 |
|---|---|---|---|---|---|
| BlazeFace FP32 | 8 | 640 | 0.915/0.892/0.797 | xx | 下载链接 |
| BlazeFace quant_post | 8 | 640 | xx/xx/xx | xx | 下载链接 |
| BlazeFace quant_aware | 8 | 640 | xx/xx/xx | xx | 下载链接 |
| BlazeFace-Lite FP32 | 8 | 640 | 0.909/0.885/0.781 | xx | 下载链接 |
| BlazeFace-Lite quant_post | 8 | 640 | xx/xx/xx | xx | 下载链接 |
| BlazeFace-Lite quant_aware | 8 | 640 | xx/xx/xx | xx | 下载链接 |
| BlazeFace-NAS FP32 | 8 | 640 | 0.837/0.807/0.658 | xx | 下载链接 |
| BlazeFace-NAS quant_post | 8 | 640 | xx/xx/xx | xx | 下载链接 |
| BlazeFace-NAS quant_aware | 8 | 640 | xx/xx/xx | xx | 下载链接 |
#
1.3 图像分割#
数据集：Cityscapes
| Model | mIoU | 模型大小（MB） | 下载 |
|---|---|---|---|
| DeepLabv3+/MobileNetv1 | 63.26 | xx | 下载链接 |
| DeepLabv3+/MobileNetv1 quant_post | xx | xx | 下载链接 |
| DeepLabv3+/MobileNetv1 quant_aware | xx | xx | 下载链接 |
| DeepLabv3+/MobileNetv2 | 69.81 | xx | 下载链接 |
| DeepLabv3+/MobileNetv2 quant_post | xx | xx | 下载链接 |
| DeepLabv3+/MobileNetv2 quant_aware | xx | xx | 下载链接 |
#
2. 剪枝#
2.1 图像分类#
数据集：ImageNet1000类
| Model | Top-1/Top-5 | 模型大小（MB） | FLOPs | 下载 |
|---|---|---|---|---|
| MobileNetV1 | 70.99%/89.68% | xx | xx | 下载链接 |
| MobileNetV1 uniform -50% | xx%/xx% | xx | xx | 下载链接 |
| MobileNetV1 sensitive -xx% | xx%/xx% | xx | xx | 下载链接 |
| MobileNetV2 | 72.15%/90.65% | xx | xx | 下载链接 |
| MobileNetV2 uniform -50% | xx%/xx% | xx | xx | 下载链接 |
| MobileNetV2 sensitive -xx% | xx%/xx% | xx | xx | 下载链接 |
| ResNet34 | 74.57%/92.14% | xx | xx | 下载链接 |
| ResNet34 uniform -50% | xx%/xx% | xx | xx | 下载链接 |
| ResNet34 auto -50% | xx%/xx% | xx | xx | 下载链接 |
#
2.2 目标检测#
数据集：Pasacl VOC & COCO 2017
| Model | 数据集 | Image/GPU | 输入608 mAP | 输入416 mAP | 输入320 mAP | 模型大小（MB） | FLOPs | 下载 |
|---|---|---|---|---|---|---|---|---|
| MobileNet-V1-YOLOv3 | Pasacl VOC | 8 | 76.2 | 76.7 | 75.3 | xx | xx | 下载链接 |
| MobileNet-V1-YOLOv3 prune xx% | Pasacl VOC | 8 | xx | xx | xx | xx | xx | 下载链接 |
| MobileNet-V1-YOLOv3 | COCO | 8 | 29.3 | 29.3 | 27.1 | xx | xx | 下载链接 |
| MobileNet-V1-YOLOv3 prune xx% | COCO | 8 | xx | xx | xx | xx | xx | 下载链接 |
| R50-dcn-YOLOv3 | COCO | 8 | 41.4 | - | - | xx | xx | 下载链接 |
| R50-dcn-YOLOv3 prune xx% | COCO | 8 | xx | - | - | xx | xx | 下载链接 |
2.3 图像分割#
数据集：Cityscapes
| Model | mIoU | 模型大小（MB） | FLOPs | 下载 |
|---|---|---|---|---|
| DeepLabv3+/MobileNetv2 | 69.81 | xx | xx | 下载链接 |
| DeepLabv3+/MobileNetv2 prune xx% | xx | xx | xx | 下载链接 |
3. 蒸馏#
3.1 图象分类#
数据集：ImageNet1000类
| Model | baseline | 下载 |
|---|---|---|
| MobileNetV1 | 70.99%/89.68% | 下载链接 |
| MobileNetV1 distilled (teacher: ResNet50_vd1) | 72.79%/90.69% | 下载链接 |
| MobileNetV2 | 72.15%/90.65% | 下载链接 |
| MobileNetV2 distilled (teacher: ResNet50_vd) | 74.30%/91.52% | 下载链接 |
| ResNet50 | 76.50%/93.00% | 下载链接 |
| ResNet50 distilled (teacher: ResNet1012) | 77.40%/93.48% | 下载链接 |
Note
[1]：ResNet50_vd预训练模型Top-1/Top-5准确率分别为79.12%/94.44%
带_vd后缀代表开启了Mixup训练，Mixup相关介绍参考mixup: Beyond Empirical Risk Minimization
3.2 目标检测#
数据集：Pasacl VOC & COCO 2017
| Model | 数据集 | Image/GPU | 输入640 mAP | 输入416 mAP | 输入320 mAP | 下载链接 |
|---|---|---|---|---|---|---|
| MobileNet-V1-YOLOv3 | Pasacl VOC | 16 | 76.2 | 76.7 | 75.3 | 下载链接 |
| MobileNet-V1-YOLOv3 distilled (teacher: ResNet34-YOLOv3-VOC3) | Pasacl VOC | 16 | xx | xx | xx | 下载链接 |
| MobileNet-V1-YOLOv3 | COCO | 16 | 29.3 | 29.3 | 27.1 | 下载链接 |
| MobileNet-V1-YOLOv3 distilled (teacher: ResNet34-YOLOv3-COCO4) | COCO | 16 | xx | xx | xx | 下载链接 |