modify README

README.md

@@ -1,6 +1,6 @@
 # Siamese and Triplet networks for image classification
 
-This repository contains implementation of a deep neural networks for embeddings learning using Siamese and Triplets approaches with different negative samples mining strategies.
+This repository contains Keras implementation of a deep neural networks for embeddings learning using Siamese and Triplets approaches with different negative samples mining strategies.
 
 # Installation
 
@@ -83,5 +83,10 @@ Is is also possible to use [test_network.ipynb](https://github.com/RocketFlash/S
 
 Result encodings could be visualized interactively using **plot_tsne_interactive** function in [utils.py](https://github.com/RocketFlash/SiameseNet/blob/master/embedding_net/utils.py).
 
-t-SNE plot of russian traffic sign images embeddings (107 classes):
+t-SNE plots of russian traffic sign images embeddings (107 classes)
+
+Before training:
+![t-SNE before](images/t-sne_without_training.png)
+
+After training:
 ![t-SNE example](images/t-sne.png)

configs/plates.yml

@@ -1,18 +1,21 @@
 input_shape :  [128, 128, 3]
-encodings_len: 1024
+encodings_len: 256
+margin: 0.5
 mode : 'triplet'
 distance_type : 'l1'
-backbone : 'resnet50'
+backbone : 'resnext50'
 backbone_weights : 'imagenet'
-optimizer : 'adam'
+optimizer : 'radam'
 learning_rate : 0.0001
 project_name : 'plates/'
 freeze_backbone : True
 augmentation_type : 'default'
+embeddings_normalization: True
 
 #paths
 dataset_path : '/home/rauf/plates_competition/dataset/to_train/'
 tensorboard_log_path : 'tf_log/'
 weights_save_path : 'weights/'
 plots_path : 'plots/'
 encodings_path : 'encodings/'
+model_save_name : 'best_model_simple2_plates.h5'

train.py

@@ -6,11 +6,11 @@
 
 
 n_epochs = 1000
-n_steps_per_epoch = 600
+n_steps_per_epoch = 50
 val_batch_size = 8
 val_steps = 100
 
-config_name = 'road_signs_resnext50_merged_dataset'
+config_name = 'plates'
 model = EmbeddingNet('configs/{}.yml'.format(config_name))
 
 initial_lr = 1e-4
@@ -34,9 +34,9 @@
                              epochs=n_epochs,
                              callbacks = callbacks, 
                              val_steps=100, 
-                             n_classes=20, 
+                             n_classes=2, 
                              n_samples=4,
-                             negative_selection_mode='hardest')
+                             negative_selection_mode='semihard')
 
 model.generate_encodings(save_file_name='encodings_{}.pkl'.format(config_name),
                          max_num_samples_of_each_classes=30, shuffle=True)