yiiwood / ademxapp

Visual applications

ademxapp

Visual applications by the University of Adelaide

For more details, refer to our report: Wider or Deeper: Revisiting the ResNet Model for Visual Recognition.

Steps to use

0. Install MXNet.

1. Download the ILSVRC 2012 classification val set, and put the extracted images into the directory:

   data/ilsvrc12/ILSVRC2012_val/

2. Download the models as below, and put them into the directory:

   models/

3. Try these commands to check the classification performance on the ILSVRC 2012 val set:

   python iclass/ilsvrc.py --data-root data/ilsvrc12 --output output --batch-images 10 --phase val --weight models/ilsvrc-cls_rna-a_cls1000_ep-0001.params --split val --test-scales 320
   
   python iclass/ilsvrc.py --data-root data/ilsvrc12 --output output --batch-images 10 --phase val --weight models/ilsvrc-cls_rna-a1_cls1000_ep-0001.params --split val --test-scales 320

Image classification

Note: Due to a change of MXNet in padding at pooling layers, some of the computed feature maps in Model A will have different sizes from as stated in our paper. However, this has no effect on Model A1, which always uses convolution layers (instead of pooling layers) for down-sampling. So, in most cases, just use Model A1, which was initialized from Model A, and further tuned for 45k extra iterations.

Results on the ILSVRC 2012 val set:

model|top-1|top-5
:---:|:---:|:---:
[Model A](https://cdn.rawgit.com/itijyou/ademxapp/master/misc/ilsvrc_model_a.pdf) [aar](https://cloudstor.aarnet.edu.au/plus/index.php/s/V7dncO4H0ijzeRj)|19.20%|4.73%
[Model A1](https://cdn.rawgit.com/itijyou/ademxapp/master/misc/ilsvrc_model_a1.pdf) [aar](https://cloudstor.aarnet.edu.au/plus/index.php/s/NOPhJ247fhVDnZH)|19.54%|4.75%

Semantic image segmentation

The below results are obtained using our pre-trained models. We will release these models soon.

Results on the PASCAL VOC test set:

model|mean IoU
:---:|:---:
Model A1, 2 conv.|82.5

Note: Model A2 was initialized from Model A, and tuned for 45k extra iterations using the Places data in ILSVRC 2016.

Results on the Cityscapes test set:

model|class IoU|class iIoU| category IoU| category iIoU
:---:|:---:|:---:|:---:|:---:
Model A2, 2 conv.|78.4|59.1|90.9|81.1
Model A2, 2 conv., coarse|79.9|59.7|91.2|80.8

Citation

If you use this code or these models in your research, please cite:

@Misc{2016.ZifengWu,
    author = {Zifeng Wu and Chunhua Shen and Anton van den Hengel},
    title = {Wider or Deeper: Revisiting the ResNet Model for Visual Recognition},
    year = {2016}
    howpublished = {arXiv:1611.10080}
}