Skip to content

Latest commit

 

History

History
84 lines (57 loc) · 3.06 KB

README.md

File metadata and controls

84 lines (57 loc) · 3.06 KB
Raw

Test Fast-RCNN on Another Dataset

We will illustrate how to test Fast-RCNN on another dataset in the following steps, and we will take INRIA Person as the example dataset.

Format Your Dataset

At first, the dataset must be well organzied with the required format.

INRIA
|-- data
    |-- Annotations
         |-- *.txt (Annotation files)
    |-- Images
         |-- *.png (Image files)
    |-- ImageSets
         |-- test.txt
|-- results
    |-- test (empty before test)
|-- VOCcode (optical)

The test.txt contains all the names(without extensions) of images files that will be used for training. For example, there are a few lines in test.txt below.

crop_000001
crop_000002
crop_000003
crop_000004
crop_000005

Construct IMDB

See it at https://github.com/EdisonResearch/fast-rcnn/tree/master/help/train.

Actually you do not need to implement the _load_inria_annotation, you could just use inria.py to construct IMDB for your own dataset. For example, to train on a dataset named TownCenter, just the followings to factory.py.

towncenter_devkit_path = '/home/szy/TownCenter'
for split in ['test']:
   name = '{}_{}'.format('towncenter', split)
   __sets[name] = (lambda split=split: datasets.inria(split, towncenter_devkit_path))

Run Selective Search

See it at https://github.com/EdisonResearch/fast-rcnn/tree/master/help/train.

Note that it should be test.mat rather than train.mat.

Modify Prototxt

For example, if you want to use the model VGG_CNN_M_1024, then you should modify test.prototxt in $FRCNN_ROOTmodels/VGG_CNN_M_1024, it mainly concerns with the number of classes you want to train. Let's assume that the number of classes is C (do not forget to count the background` class). Then you should

  • Modify num_output in the cls_score layer to C
  • Modify num_output in the bbox_pred layer to 4 * C

See rbgirshick#11 for more details.

Prepare Your Evaluation Code

In the original framework of Fast-RCNN, it uses matlab wrappers to evluate the results. As the evluation process is not very difficult, you could modify the function evaluate_detections in inria.py.

As INRIA Person provides some matlab files in the format of PASCAL-VOC, you could modify it a little and use it directly. You could see https://github.com/EdisonResearch/fast-rcnn/tree/master/help/INRIA/VOCcode for the VOCcode.

If you do not want to use the evluation function in the framework of Fast-RCNN, you could find the results in the directory results/test in the roor directory of your dataset.

Test!

In the directory $FRCNN_ROOT, run the following command in the shell.

./tools/test_net.py --gpu 1 --def models/VGG_CNN_M_1024/test.prototxt \
    --net output/default/train/vgg_cnn_m_1024_fast_rcnn_iter_40000.caffemodel --imdb inria_test

Be careful with the imdb argument as it specifies the dataset you will train on.

References

[Fast-RCNN] https://github.com/rbgirshick/fast-rcnn

Contact

Feel free to contact me at zeyuanxy@gmail.com.