Skip to content

Latest commit

 

History

History
134 lines (109 loc) · 33.2 KB

README-model-zoo.md

File metadata and controls

134 lines (109 loc) · 33.2 KB
Raw

Model Zoo

⚠️ For recent papers along with pre-trained models, training/evaluation recipes, and configuration files, please see examples folder. We will update model zoo periodically.⚠️

This file contains the links to all the pre-trained models in CVNets and their configs:

Classification (ImageNet-1k)

Model Parameters Top-1 Pretrained weights Config file Logs
ViT-tiny 5.7 M 72.91 Link Link Link
ResNet-34 21.8 M 74.85 Link Link Link
ResNet-50 25.6 M 78.44 Link Link Link
ResNet-101 44.5 M 79.81 Link Link Link
MobileNetv1-0.25 0.5 M 54.45 Link Link Link
MobileNetv1-0.5 1.3 M 65.93 Link Link Link
MobileNetv1-0.75 2.6 M 71.44 Link Link Link
MobileNetv1-1.00 4.2 M 74.04 Link Link Link
MobileNetv2-0.25 1.5 M 53.57 Link Link Link
MobileNetv2-0.5 2.0 M 65.28 Link Link Link
MobileNetv2-0.75 2.6 M 70.42 Link Link Link
MobileNetv2-1.00 3.5 M 72.93 Link Link Link
MobileNetv3-small 2.5 M 66.65 Link Link Link
MobileNetv3-large 5.4 M 75.13 Link Link Link
ResNet-34 (advanced recipe) 21.8 M 76.91 Link Link Link
ResNet-50 (advanced recipe) 25.6 M 80.36 Link Link Link
ResNet-101 (advanced recipe) 44.5 M 81.68 Link Link Link

MobileViTv1 (Legacy)

Note: These resutls are from CVNets v0.1. We discontinued the support of OpenCV and switched to PIL in v0.2. For MobileViTv1 results, see v0.1.

Model Parameters Top-1 Pretrained weights Config file
MobileViT-XXS 1.3 M 69.0 Link Link
MobileViT-XS 2.3 M 74.7 Link Link
MobileViT-S 5.6 M 78.3 Link Link

MobileViTv2 (256x256)

Model Parameters Top-1 Pretrained weights Config file Logs
MobileViTv2-0.5 1.4 M 70.18 Link Link Link
MobileViTv2-0.75 2.9 M 75.56 Link Link Link
MobileViTv2-1.0 4.9 M 78.09 Link Link Link
MobileViTv2-1.25 7.5 M 79.65 Link Link Link
MobileViTv2-1.5 10.6 M 80.38 Link Link Link
MobileViTv2-1.75 14.3 M 80.84 Link Link Link
MobileViTv2-2.0 18.4 M 81.17 Link Link Link

MobileViTv2 (Trained on 256x256 and Finetuned on 384x384)

Model Parameters Top-1 Pretrained weights Config file Logs
MobileViTv2-0.5 1.4 M 72.14 Link Link Link
MobileViTv2-0.75 2.9 M 76.98 Link Link Link
MobileViTv2-1.0 4.9 M 79.68 Link Link Link
MobileViTv2-1.25 7.5 M 80.94 Link Link Link
MobileViTv2-1.5 10.6 M 81.50 Link Link Link
MobileViTv2-1.75 14.3 M 82.04 Link Link Link
MobileViTv2-2.0 18.4 M 82.17 Link Link Link

MobileViTv2 (Trained on ImageNet-21k and Finetuned on ImageNet-1k 256x256)

Model Parameters Top-1 Pretrained weights Config file Logs
MobileViTv2-1.5 10.6 M 81.46 Link Link Link
MobileViTv2-1.75 14.3 M 81.94 Link Link Link
MobileViTv2-2.0 18.4 M 82.36 Link Link Link

MobileViTv2 (Trained on ImageNet-21k, Finetuned on ImageNet-1k 256x256, and Finetuned on ImageNet-1k 384x384)

Model Parameters Top-1 Pretrained weights Config file Logs
MobileViTv2-1.5 10.6 M 82.60 Link Link Link
MobileViTv2-1.75 14.3 M 82.93 Link Link Link
MobileViTv2-2.0 18.4 M 83.41 Link Link Link

Object Detection (MS-COCO)

Model Parameters MAP Pretrained weights Config file Logs
SSD ResNet-50 28.5 M 29.98 Link Link Link
SSD MobileViTv2-0.5 2.0 M 21.24 Link Link Link
SSD MobileViTv2-0.75 3.6 M 24.57 Link Link Link
SSD MobileViTv2-1.0 5.6 M 26.47 Link Link Link
SSD MobileViTv2-1.25 8.2 M 27.85 Link Link Link
SSD MobileViTv2-1.5 11.3 M 28.83 Link Link Link
SSD MobileViTv2-1.75 14.9 M 29.52 Link Link Link
SSD MobileViTv2-2.0 19.1 M 30.21 Link Link Link

Segmentation

Note: The number of parameters reported does not include the auxiliary branches.

ADE20K Dataset

Model Parameters mIoU Pretrained weights Config file Logs
DeepLabv3 MobileNetv2 8.0 M 35.20 Link Link Link
PSPNet MobileViTv2-0.5 3.6 M 31.77 Link Link Link
PSPNet MobileViTv2-0.75 6.2 M 35.22 Link Link Link
PSPNet MobileViTv2-1.0 9.4 M 36.57 Link Link Link
PSPNet MobileViTv2-1.25 13.2 M 38.76 Link Link Link
PSPNet MobileViTv2-1.5 17.6 M 38.74 Link Link Link
PSPNet MobileViTv2-1.75 22.5 M 39.82 Link Link Link
DeepLabv3 MobileViTv2-0.5 6.3 M 31.93 Link Link Link
DeepLabv3 MobileViTv2-0.75 9.6 M 34.70 Link Link Link
DeepLabv3 MobileViTv2-1.0 13.4 M 37.06 Link Link Link
DeepLabv3 MobileViTv2-1.25 17.7 M 38.42 Link Link Link
DeepLabv3 MobileViTv2-1.5 22.6 M 38.91 Link Link Link
DeepLabv3 MobileViTv2-1.75 28.1 M 39.53 Link Link Link
DeepLabv3 MobileViTv2-2.0 34.0 M 40.94 Link Link Link

Pascal VOC 2012 Dataset

Model Parameters mIoU Pretrained weights Config file Logs
DeepLabv3 MobileViTv1 8.5 M 79.44 Link Link Link
PSPNet MobileViTv2-0.5 3.6 M 74.62 Link Link Link
PSPNet MobileViTv2-0.75 6.2 M 77.44 Link Link Link
PSPNet MobileViTv2-1.0 9.4 M 78.92 Link Link Link
PSPNet MobileViTv2-1.25 13.2 M 79.40 Link Link Link
PSPNet MobileViTv2-1.5 17.5 M 79.93 Link Link Link
DeepLabv3 MobileViTv2-0.5 6.2 M 75.07 Link Link Link
DeepLabv3 MobileViTv2-1.0 13.3 M 78.94 Link Link Link
DeepLabv3 MobileViTv2-1.25 17.7 M 79.68 Link Link Link
DeepLabv3 MobileViTv2-1.5 22.6 M 80.30 Link Link Link

Video Classification (Kinetics-400)

Model Parameters Top-1 Pretrained weights Config file Logs
MobileViTv1-small-SpatioTemporal 5.2 M 68.38 Link Link Link