Attribute Based Learning implementation

Authors : Charles Corbière, Bernardo Cardoso Cordeiro, Aymeric Zhuo, Luciano Di Palma

Synopsis

This package implement differents attribute-based learning approach:

• Direct Attributed Prediction (iAP)
• Indirect Attributed Prediction (IAP)

For this matter, we choose to try two learner :

• SVM
• Neural Network

Dataset included:

• Animals with Attributed dataset
• PubFig

For AwA dataset, how to use it

0. Download Animal VGG19 features here and decompress it.

1. Compute features into animal categories in ./feat folder

python createFeaturesVector.py pathToFeatures

2. Concatenate features into a training dataset and a test dataset with their respectives labels in ./CreatedDataset

python concatenateSetFeatures.py

3. Train and predict a model given a method and a classifier. Platt parameters, prediction and probabilities saved in ./DAP

python DirectAttributePrediction.py DAP SVM

3. (bis) Train and predict using IAP with SVM. Prediction and probabilities saved in ./IAP

python indirectAttributePrediction.py IAP

4. Generate and save confusion matrix plot and roc in ./results given a classifier

python DAP_eval.py SVM

4. (bis) Generate and save confusion matrix plot and roc in ./results given a classifier

python IAP_eval.py

For Pubfig dataset, how to use it

0. Download Animal VGG19 features here and decompress it on Datasets/PubFig.

1. Compute attributes_names, celebrities and attributes as an array

python preparePubFigFiles.py