add constants file for gpu-3 machine

constants.py

@@ -5,6 +5,8 @@ if hostname=='vision-gpu-1':
     from constants_vision_gpu_1 import *
 elif hostname=='vision-gpu-2':
     from constants_vision_gpu_2 import *
+elif hostname=='vision-gpu-3':
+    from constants_vision_gpu_3 import *
 elif hostname=='crunchy':
     from constants_crunchy import *
 

constants_vision_gpu_3.py

+import os
+import pdb
+
+def mkdir_if_not_exists(dir_name):
+    if not os.path.exists(dir_name):
+        os.mkdir(dir_name)
+        
+experiment_name = 'ans_through_obj_atr_det_debug_rel'
+
+##########################################################################
+#                    Machine Specific Paths                              #
+##########################################################################
+
+# Global output directory (all subexperiments will be saved here)
+global_output_dir = '/data/tanmay/GenVQA_Exp_Results'
+
+global_experiment_dir = os.path.join(
+    global_output_dir,
+    experiment_name)
+
+tb_log_dir = os.path.join(
+    global_experiment_dir,
+    'tensorboard_logdir')
+
+mkdir_if_not_exists(global_output_dir)
+mkdir_if_not_exists(global_experiment_dir)
+mkdir_if_not_exists(tb_log_dir)
+
+# Genome Parent Directory
+data_absolute_path = '/home/nfs/tgupta6/data/VisualGenome'
+
+# Pretrained resnet ckpt
+resnet_ckpt = '/home/nfs/tgupta6/data/Resnet/' + \
+              'ResNet-L50.ckpt'
+
+# Pretrained word vectors
+word2vec_binary = '/home/nfs/tgupta6/data/word_vectors/' + \
+                  'GoogleNews-vectors-negative300.bin'
+
+# VQA parent directory
+vqa_basedir = '/home/nfs/tgupta6/data/VQA/'
+
+# Pretrained obj atr model to be restored
+pretrained_model = '/home/nfs/tgupta6/projects/GenVQA/Exp_Results/' +\
+                   'object_attribute_classifier_large_images_vqa_split/' +\
+                   'object_attribute_classifiers/model-80000'
+
+##########################################################################
+#                         Model Parameters                               #
+##########################################################################
+
+# height and width to which images are resized before feeding into networks
+image_size = (224, 224) 
+
+# Token to be used if object or attribute variable is unknown
+unknown_token = 'UNK'
+
+##########################################################################
+#                  Labels and Word2Vec Parameters                        #
+##########################################################################
+
+num_object_labels = 1000
+num_attribute_labels = 1000
+
+word_vector_size = 300
+resnet_feat_dim = 2048
+
+# Numpy matrix storing vocabulary word vectors
+pretrained_vocab_word_vectors_npy = os.path.join(
+    data_absolute_path,
+    'restructured/pretrained_vocab_word_vectors.npy')
+
+##########################################################################
+#                         Genome Parameters                              #
+##########################################################################
+
+# Genome paths
+image_dir = os.path.join(data_absolute_path, 'cropped_regions_large')
+genome_resnet_feat_dir = os.path.join(
+    data_absolute_path,
+    'cropped_regions_large_resnet_features')
+
+object_labels_json = os.path.join(
+    data_absolute_path,
+    'restructured/object_labels.json')
+
+attribute_labels_json = os.path.join(
+    data_absolute_path,
+    'restructured/attribute_labels.json')
+
+regions_json = os.path.join(
+    data_absolute_path,
+    'restructured/region_with_hypernym_labels.json')
+
+mean_image_filename = os.path.join(
+    data_absolute_path,
+    'restructured/mean_image.jpg')
+
+vocab_json = os.path.join(
+    data_absolute_path,
+    'restructured/vocab_subset.json')
+
+genome_train_region_ids = os.path.join(
+    data_absolute_path,
+    'restructured/train_region_ids_simple.json')
+
+genome_val_region_ids = os.path.join(
+    data_absolute_path,
+    'restructured/val_region_ids_simple.json')
+
+genome_test_region_ids = os.path.join(
+    data_absolute_path,
+    'restructured/test_region_ids_simple.json')
+
+##########################################################################
+#               Genome Training/Fine-Tuning Parameters                   #
+##########################################################################
+
+# Logging parameters
+region_log_every_n_iter = 500
+region_output_dir = os.path.join(
+    global_experiment_dir,
+    'object_attribute_classifiers')
+    
+mkdir_if_not_exists(region_output_dir)
+
+region_model = os.path.join(
+    region_output_dir,
+    'model')                    
+
+# Object Attribute Classifier Training Params
+region_batch_size = 200
+region_num_epochs = 20
+region_queue_size = 400
+region_regularization_coeff = 1e-5
+region_lr = 1e-3
+
+# Object Attribute Finetuning Params
+region_fine_tune_from_iter = 50500
+region_fine_tune_from = region_model + '-' + str(region_fine_tune_from_iter)
+
+##########################################################################
+#                     Genome Evaluation Parameters                       #
+##########################################################################
+
+# Object Attribute Model Selection
+region_start_model = 8000
+region_step_size = 8000
+region_model_accuracies_txt = os.path.join(
+    region_output_dir,
+    'model_accuracies.txt')
+
+# Object Attribute Classifier Evaluation Params
+region_eval_on = 'train_held_out' # One of {'test','train_held_out','train_subset'}
+region_model_to_eval = region_model + '-' + '102000'
+
+# Path to results
+visualize_object_predictions = False
+
+region_pred_vis_dirname = os.path.join(
+    region_output_dir,
+    'region_pred_vis_'+region_eval_on)
+
+mkdir_if_not_exists(region_pred_vis_dirname)
+
+region_attribute_scores_dirname = os.path.join(
+    region_output_dir,
+    'attribute_scores')
+
+region_object_scores_dirname = os.path.join(
+    region_output_dir,
+    'object_scores')
+
+mkdir_if_not_exists(region_attribute_scores_dirname)
+mkdir_if_not_exists(region_object_scores_dirname)
+
+##########################################################################
+#                         VQA Parameters                                 #
+##########################################################################
+
+# Answer prediction
+num_region_proposals = 100
+num_mcq_candidates = 18
+num_negative_answers = num_mcq_candidates - 1
+
+# VQA data paths
+vqa_train_image_dir = os.path.join(
+    vqa_basedir,
+    'train2014_cropped_large')
+vqa_train_resnet_feat_dir = os.path.join(
+    vqa_basedir,
+    'train2014_cropped_large_resnet_features')
+vqa_train_anno = os.path.join(
+    vqa_basedir,
+    'mscoco_train2014_annotations_with_parsed_questions.json')
+vqa_train_subset_qids = os.path.join(
+    vqa_basedir,
+    'train_subset_qids.json')
+vqa_train_held_out_qids = os.path.join(
+    vqa_basedir,
+    'train_held_out_qids.json')
+
+vqa_val_image_dir = os.path.join(
+    vqa_basedir,
+    'val2014_cropped_large')
+vqa_val_resnet_feat_dir = os.path.join(
+    vqa_basedir,
+    'val2014_cropped_large_resnet_features')
+vqa_val_anno = os.path.join(
+    vqa_basedir,
+    'mscoco_val2014_annotations_with_parsed_questions.json')
+vqa_val_qids = os.path.join(
+    vqa_basedir,
+    'val_qids.json')
+
+vqa_answer_vocab_json = os.path.join(
+    vqa_basedir,
+    'answer_vocab.json')
+
+# Dataset statistics
+# num_train_questions = 248349
+# num_train_held_out_questions = 12500
+# num_train_subset_questions = num_train_questions - num_train_held_out_questions
+# num_val_questions = 121512
+
+##########################################################################
+#                   VQA Training/Finetuning Parameters                   #
+##########################################################################
+
+# Logging parameters
+answer_log_every_n_iter = 500
+answer_output_dir = os.path.join(
+    global_experiment_dir,
+    'answer_classifiers')
+    
+mkdir_if_not_exists(answer_output_dir)
+
+answer_model = os.path.join(
+    answer_output_dir,
+    'model')
+
+# Answer classifier training params
+answer_train_from_scratch = True
+
+answer_batch_size = 25
+answer_num_epochs = 20
+answer_queue_size = 500
+answer_regularization_coeff = 1e-5
+answer_lr = 1e-3
+
+answer_obj_atr_loss_wt = 0.1
+answer_ans_loss_wt = 1.0
+answer_mil_loss_wt = 0.0
+
+# Answer classifier additional joint training params
+num_regions_with_labels = 100
+
+# Answer fine tune params
+answer_fine_tune_from_iter = 29500
+answer_fine_tune_from = answer_model + '-' + str(answer_fine_tune_from_iter)
+
+##########################################################################
+#                        VQA Evaluation Parameters                       #
+##########################################################################
+
+# Select best model
+models_dir = answer_output_dir
+start_model = 30000
+step_size = 4000
+model_accuracies_txt = os.path.join(
+    answer_output_dir,
+    'model_accuracies.txt')
+
+# Answer eval params
+answer_eval_on = 'val'
+answer_model_to_eval = answer_model + '-66000'
+
+vqa_results_dir = os.path.join(
+    answer_output_dir,
+    'Results')
+
+mkdir_if_not_exists(vqa_results_dir)
+
+answer_eval_data_json = os.path.join(
+    vqa_results_dir,
+    'eval_' + answer_eval_on + '_data.json')
+
+answer_eval_results_json = os.path.join(
+    vqa_results_dir,
+    'eval_' + answer_eval_on + '_results.json')
+
+# Fine Grained Evaluation File paths
+raw_vqa_val_ques_json = os.path.join(
+    vqa_basedir,
+    'MultipleChoice_mscoco_val2014_questions.json')
+
+raw_vqa_val_anno_json = os.path.join(
+    vqa_basedir,
+    'mscoco_val2014_annotations.json')