def train():
dataset = VidVRD('../vidvrd-dataset', '../vidvrd-dataset/videos', ['train', 'test'])
param = dict()
param['model_name'] = 'baseline'
param['rng_seed'] = 1701
param['max_sampling_in_batch'] = 32
param['batch_size'] = 64
param['learning_rate'] = 0.001
param['weight_decay'] = 0.0
param['max_iter'] = 5000
param['display_freq'] = 1
param['save_freq'] = 5000
param['epsilon'] = 1e-8
param['pair_topk'] = 20
param['seg_topk'] = 200
print(param)
model.train(dataset, param)
def train():
dataset = VidVRD(anno_rpath=anno_rpath,
video_rpath=video_rpath,
splits=splits)
param = dict()
param['model_name'] = 'baseline'
param['rng_seed'] = 1701
param['max_sampling_in_batch'] = 32
param['batch_size'] = 64
param['learning_rate'] = 0.001
param['weight_decay'] = 0.0
param['max_iter'] = 5000
param['display_freq'] = 1
param['save_freq'] = 5000
param['epsilon'] = 1e-8
param['pair_topk'] = 20
param['seg_topk'] = 200
print(param)
model.train(dataset, param)
def train_epoc(trn,model,criterion,optimizer,scheduler):
# Train the model
model.train()
train_loss = 0
train_acc = 0
for text_ap, offsets_ap, text_cid,offsets_cid, clss in tqdm(trn):
optimizer.zero_grad()
text_ap, text_cid=text_ap.long().to(device),text_cid.long().to(device)
offsets_ap, offsets_cid, clss = offsets_ap.to(device),offsets_cid.to(device), clss.to(device)
output = model(text_ap, offsets_ap,text_cid, offsets_cid)
loss = criterion(output, clss)
train_loss += loss.item()
loss.backward()
optimizer.step()
train_acc += (output.argmax(1) == clss).sum().item()
# Adjust the learning rate
scheduler.step()
return train_loss , train_acc
