single TensorFlow .pb files

Freeze and convert trained Keras models into a single TensorFlow pb file.

Bottleneck features

Saved bottleneck features of various pre-trained networks

Data of Human Rights Archive

There are 3050 train images from 8 human rights violations categories and one 'no violation' category in the HRA, which are used to train the Human-Rights-Archive-CNNs.
The validation set can automatically be set in Keras by setting validation_split argument in model.fit accordingly. There are 30 images per category in the testing set.

CompoundNet feature extraction weights files

Add feature extraction mode weights for CompoundNet

batch_size = 25
feature_extraction_epochs = 10

class_weight = {0: 5.08, 1: 1, 2: 10.86, 3: 5.08, 4: 3.46, 5: 2.31, 6: 4.70, 7: 6.17, 8: 1.55}

Fine-tuning weights files

Add fine-tuning mode weights for VGG16, VGG16-Places365 and ResNet50.

batch_size = 25
fine_tune_epochs = 20

class_weight = {0: 5.08, 1: 1, 2: 10.86, 3: 5.08, 4: 3.46, 5: 2.31, 6: 4.70, 7: 6.17, 8: 1.55}

Feature extraction weights files

Add feature extraction mode weights for VGG16, VGG16-Places365 and ResNet50.

batch_size = 25
feature_extraction_epochs = 10

class_weight = {0: 5.08, 1: 1, 2: 10.86, 3: 5.08, 4: 3.46, 5: 2.31, 6: 4.70, 7: 6.17, 8: 1.55}

Baseline model weights files

Add weights files for the baseline (trained from scratch) HRA-CNN model

_________________________________________________________________
Layer (type)                 Output Shape              Param #
=================================================================
input_1 (InputLayer)         (None, 224, 224, 3)       0
_________________________________________________________________
block1_conv1 (Conv2D)        (None, 224, 224, 32)      896
_________________________________________________________________
block1_pool (MaxPooling2D)   (None, 112, 112, 32)      0
_________________________________________________________________
block2_conv1 (Conv2D)        (None, 112, 112, 32)      9248
_________________________________________________________________
block2_pool (MaxPooling2D)   (None, 56, 56, 32)        0
_________________________________________________________________
block3_conv1 (Conv2D)        (None, 56, 56, 64)        18496
_________________________________________________________________
block3_pool (MaxPooling2D)   (None, 28, 28, 64)        0
_________________________________________________________________
flatten (Flatten)            (None, 50176)             0
_________________________________________________________________
fc1 (Dense)                  (None, 64)                3211328
_________________________________________________________________
dropout_1 (Dropout)          (None, 64)                0
_________________________________________________________________
predictions (Dense)          (None, 9)                 585
=================================================================
Total params: 3,240,553
Trainable params: 3,240,553
Non-trainable params: 0
_________________________________________________________________