Authors : Charles Corbière, Bernardo Cardoso Cordeiro, Aymeric Zhuo, Luciano Di Palma
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
-
Download Animal VGG19 features here and decompress it.
-
Compute features into animal categories in ./feat folder
python createFeaturesVector.py pathToFeatures
- Concatenate features into a training dataset and a test dataset with their respectives labels in ./CreatedDataset
python concatenateSetFeatures.py
- Train and predict a model given a method and a classifier. Platt parameters, prediction and probabilities saved in ./DAP
python DirectAttributePrediction.py DAP SVM
- (bis) Train and predict using IAP with SVM. Prediction and probabilities saved in ./IAP
python indirectAttributePrediction.py IAP
- Generate and save confusion matrix plot and roc in ./results given a classifier
python DAP_eval.py SVM
- (bis) Generate and save confusion matrix plot and roc in ./results given a classifier
python IAP_eval.py
-
Download Animal VGG19 features here and decompress it on Datasets/PubFig.
-
Compute attributes_names, celebrities and attributes as an array
python preparePubFigFiles.py