def test_colorization(self): imgs_l, imgs_true_ab, imgs_emb = self._tensors() # Build the network and the optimizer step col = Colorization(256) imgs_ab = col.build(imgs_l, imgs_emb) opt_operations = color_optimizer(imgs_ab, imgs_true_ab) self._run(imgs_l, imgs_ab, imgs_true_ab, opt_operations)
def test_colorization(self): imgs_l, imgs_true_ab, imgs_emb = self._tensors() # Build the network and the optimizer step col = Colorization(256) imgs_ab = col.build(imgs_l, imgs_emb) cost = tf.reduce_mean(tf.squared_difference(imgs_ab, imgs_true_ab)) optimizer = tf.train.AdamOptimizer(0.001).minimize(cost) opt_operations = {'cost': cost, 'optimizer': optimizer} self._run(imgs_l, imgs_ab, imgs_true_ab, opt_operations)
outdir = "/opt/data_out/test" used_model = 67999 if __name__ == "__main__": device = torch.device("cuda" if torch.cuda.is_available() else "cpu") davis_path = "/opt/data_davis" trainset = DavisDataset(base_path=davis_path) trainloader = torch.utils.data.DataLoader(trainset, batch_size=1, shuffle=False, num_workers=0) #Create network model = Colorization() #load model model.load_state_dict(torch.load(model_path)) # model.to(device) #Set model to eval mode. model.eval() #Read from data for i, data in enumerate(trainloader, 0): # get the inputs; data is a list of [inputs, labels] # Inputs is normalized 256,256 grayscale image: 2, 1, 256, 256 # Labels is annotations: 2, 1, 256, 256 inputs, originals, labels_dummy, key = data inputs = [x.squeeze(0) for x in inputs] originals = [x.squeeze(0) for x in originals] labels_dummy = [x.type(torch.int64) for x in labels_dummy]
epochs = 200 #default 100 val_number_of_images = 10 total_train_images = 65000 #default 130 * 500 batch_size = 38 #default 100 learning_rate = 0.0001 #default 0.001 batches = total_train_images // batch_size # START print_term('Starting session...', run_id) sess = tf.Session() K.set_session(sess) print_term('Started session...', run_id) # Build the network and the various operations print_term('Building network...', run_id) col = Colorization(256) fwd_col = Feedforward_Colorization(256) ref = Refinement() opt_operations = training_pipeline(col, fwd_col, ref, learning_rate, batch_size) evaluations_ops = evaluation_pipeline(col, fwd_col, ref, val_number_of_images) train_col_writer, train_fwd_writer, train_ref_writer, val_col_writer, val_fwd_writer, val_ref_writer = metrics_system( run_id, sess) saver, checkpoint_paths, latest_checkpoint = checkpointing_system(run_id) print_term('Built network', run_id) with sess.as_default(): # tf.summary.merge_all() # writer = tf.summary.FileWriter('./graphs', sess.graph)
from torchvision import transforms from labels_viewer import GenerateLabelImage import cv2 import os import matplotlib.pyplot as plt import numpy as np if __name__ == "__main__": device = torch.device("cuda" if torch.cuda.is_available() else "cpu") kinetics_path = "/opt/data" trainset = KineticsClustered(base_path=kinetics_path) trainloader = torch.utils.data.DataLoader(trainset, batch_size=32, shuffle=False, num_workers=0) model = Colorization() # model.load_state_dict(torch.load("/opt/model/8/models/model{}.pth".format(65999))) # print(model) model.setdev(device) parallel_model = nn.DataParallel(model, device_ids=[0, 1]) parallel_model.to(device) criterion = nn.CrossEntropyLoss() # optimizer = optim.SGD(model.parameters(), lr=0.01, momentum=0.9) optimizer = optim.Adam(parallel_model.parameters(), lr=0.001) loss_values = [] for epoch in range(10): # loop over the dataset multiple times running_loss = 0.0