Validation images in batches.

[AYG-DL]

Hi,

I am trying to send a batch of 64 images and trying to extract the features all at once from the model. However, after a certain number of images I get erroneous image features. Any idea why could that be?
I modified the extract_features.py file to do it.

[AlfredXiangWu]

I have no idea about it. Could you show me your modified code?

[AYG-DL]

def main(image,batchSize):
timer0 = time.time()
model = LightCNN_9Layers(num_classes=79077)
model.eval()
model = torch.nn.DataParallel(model).cuda()
checkpoint = torch.load("/scratch/user/ayu2224/CV/De-Occlude/dcgan_code_files/LightCNN/LightCNN_9Layers_checkpoint.pth.tar")
model.load_state_dict(checkpoint['state_dict'])
timer1 = time.time()
transform = transforms.Compose([transforms.ToTensor()])
count = 0
input = torch.zeros(batchSize, 1, 128, 128)
#image = image[:,0,:,:]
input = input.cuda()
image.resize_as_(input)
input = image
#print("")
#print(type(input))
#print("")
input_var = torch.autograd.Variable(input, volatile=True)
_, features = model(input_var)
timer2 = time.time()
print ("Checkpoint: ", timer1 -timer0)
print ("Model: ", timer2-timer1)
#print("coming here")
#print(type(features))
#print(features.data.size())
#print(type(features.data))
return features.data.cuda()

//here is the snippet of the code that I am using. The batchsize that I used was 64. I load the pretrained model provided by you.

[AYG-DL]

I really would like advice on this as I have a deadline to meet.