Testing on single image

[adithyagaurav]

Hi, I am trying to implement your model with the imagenet pre-trained weights you have provided on the repository. I'm hoping to run inference on a single image. The problem I'm facing is that, every time I run inference, model gives the output tensor(600), which means it's predicting the class for every image to be 600. I have tried different images (of different classes), the model consistently labels every image to 600.

I wish to know why must this be happening, am I doing something wrong? Following is my code:

model = darknet53(1000)
checkpoint = torch.load(checkpoint_path, map_location=torch.device('cpu'))
model.load_state_dict(checkpoint['state_dict'])

test_transform = transforms.Compose([
            transforms.Resize(256),
            transforms.CenterCrop(224),
            transforms.ToTensor(),
            transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
        ])

pil_image = test_transform(Image.open(image_path))
print(pil_image.shape)
torch_image = pil_image.unsqueeze(0)
print(torch_image.shape)

out = model(torch_image)
label = torch.argmax(out)
print(label)

Can you help me?

[ITBoy-China]

Hi, I am trying to implement your model with the imagenet pre-trained weights you have provided on the repository. I'm hoping to run inference on a single image. The problem I'm facing is that, every time I run inference, model gives the output tensor(600), which means it's predicting the class for every image to be 600. I have tried different images (of different classes), the model consistently labels every image to 600.

I wish to know why must this be happening, am I doing something wrong? Following is my code:

model = darknet53(1000)
checkpoint = torch.load(checkpoint_path, map_location=torch.device('cpu'))
model.load_state_dict(checkpoint['state_dict'])

test_transform = transforms.Compose([
            transforms.Resize(256),
            transforms.CenterCrop(224),
            transforms.ToTensor(),
            transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
        ])

pil_image = test_transform(Image.open(image_path))
print(pil_image.shape)
torch_image = pil_image.unsqueeze(0)
print(torch_image.shape)

out = model(torch_image)
label = torch.argmax(out)
print(label)

Can you help me?

You forgot to set 'model.eval()'.