def get_correctness():
model.eval()
correct_count = 0
total_count = 0
print('Testing:')
for val_batch in tqdm(val_loader):
data = val_batch
inputs = data.get('inputs')
current_batch_size = inputs.shape[0]
out = model(data, device)
if input_type == 'depth_pointcloud':
out = out[0]
class_gt = data.get('category').to(device)
class_predict = out.max(dim=1)[1]
#print('gt:',class_gt)
#print('pred:',class_predict)
correct_count += (class_predict == class_gt).sum().item()
total_count += current_batch_size
print('correct_count:', correct_count)
print('total_count:', total_count)
return correct_count / total_count
def forward(self, data, device):
gt_depth_maps = data.get('inputs.depth').to(device)
gt_mask = data.get('inputs.mask').to(device).byte()
background_setting(gt_depth_maps, gt_mask)
encoder_inputs = gt_depth_maps
if self.with_img:
img = data.get('inputs').to(device)
encoder_inputs = {'img': img, 'depth': encoder_inputs}
out = self.features(encoder_inputs)
out = self.pred_fc(out)
return out
def get_loss(self, data, device):
class_gt = data.get('category').to(device)
out = self.forward(data, device)
loss = self.loss_func(out, class_gt)
return loss
def forward(self, data, device):
inputs = data.get('inputs').to(device)
out = self.features(inputs)
out = self.fc(out)
return out