| name | caffemodel | prototxt | training set |
|---|---|---|---|
| CurriculumNet(WebVision) | inception_v2_webvision.caffemodel | inception_v2_deploy_webvision.prototxt | WebVision 1.0 |
| CurriculumNet(WebVision+ImageNet) | inception_v2_web_img.caffemodel | inception_v2_deploy_web_img.prototxt | WebVision 1.0 + ImageNet |
These models are trained using the strategy described in the CurriculumNet paper.
The performance on the validation sets of ImageNet and WebVision are as follows:
| Name | WebVision (Top-1) | WebVision (Top-5) | ImageNet (Top-1) | ImageNet (Top-5) |
|---|---|---|---|---|
| CurriculumNet(WebVision) | 27.9 | 10.8 | 35.2 | 15.1 |
| CurriculumNet(WebVision+ImageNet) | 24.7 | 8.5 | 24.8 | 7.1 |
The deep learning framework is a fork of Caffe with OpenMPI-based Multi-GPU: https://github.com/yjxiong/caffe/
The models are released for non-commercial use. For commercial use, please contact [email protected]