Get the network input dimensions of the pretrained model

When learning about neural networks,Pre-trained network packages are often used.
例如

from torchvision import models
resnet = models.res18(pretrained=True)

But when we are beginners in the use of puzzled：What size image do I need to enter?？
解决方案：

方法一：读torchvision.models的说明文档

打开torchvision.models的网站：
https://pytorch.org/hub/research-models

Search for the model name you need,得到resnet的网站：
https://pytorch.org/hub/pytorch_vision_resnet/
在里面看到resnet的指导文档：

import torch
model = torch.hub.load('pytorch/vision:v0.10.0', 'resnet18', pretrained=True)
# or any of these variants
# model = torch.hub.load('pytorch/vision:v0.10.0', 'resnet34', pretrained=True)
# model = torch.hub.load('pytorch/vision:v0.10.0', 'resnet50', pretrained=True)
# model = torch.hub.load('pytorch/vision:v0.10.0', 'resnet101', pretrained=True)
# model = torch.hub.load('pytorch/vision:v0.10.0', 'resnet152', pretrained=True)
model.eval()

All pre-trained models expect input images normalized in the same way, i.e. mini-batches of 3-channel RGB images of shape (3 x H x W), where H and W are expected to be at least 224. The images have to be loaded in to a range of [0, 1] and then normalized using mean = [0.485, 0.456, 0.406] and std = [0.229, 0.224, 0.225].
Here’s a sample execution.

# sample execution (requires torchvision)
from PIL import Image
from torchvision import transforms
input_image = Image.open(filename)
preprocess = 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]),
])
input_tensor = preprocess(input_image)
input_batch = input_tensor.unsqueeze(0) # create a mini-batch as expected by the model

# move the input and model to GPU for speed if available
if torch.cuda.is_available():
    input_batch = input_batch.to('cuda')
    model.to('cuda')

with torch.no_grad():
    output = model(input_batch)
# Tensor of shape 1000, with confidence scores over Imagenet's 1000 classes
print(output[0])
# The output has unnormalized scores. To get probabilities, you can run a softmax on it.
probabilities = torch.nn.functional.softmax(output[0], dim=0)
print(probabilities)

从中,可以了解到,The image needs to be scaled to size[224, 224].

方法二：读GitHub源代码

除了看torchvision的指导文档,也可以进入githubsee the comments in the source code(View on GitHub)：

GithubThe code includes the following：

class ResNet18_Weights(WeightsEnum):
    IMAGENET1K_V1 = Weights(
        url="https://download.pytorch.org/models/resnet18-f37072fd.pth",
        transforms=partial(ImageClassification, crop_size=224),
        meta={
    
            **_COMMON_META,
            "num_params": 11689512,
            "recipe": "https://github.com/pytorch/vision/tree/main/references/classification#resnet",
            "_metrics": {
    
                "ImageNet-1K": {
    
                    "[email protected]": 69.758,
                    "[email protected]": 89.078,
                }
            },
            "_docs": """These weights reproduce closely the results of the paper using a simple training recipe.""",
        },
    )
    DEFAULT = IMAGENET1K_V1

由其中的crop_size=224Know that it should be scaled or cropped to224大小.

方法三：暴力测试

The size of the size design for yourself think feasible,loop in a certain interval,如果尺寸不合适,PytorchIt will report an error that the model size is not compatible.用exceptSkip these exceptions,保留tryThe size of the success.

transfer_model = resnet18(pretrained=True)
print(transfer_model)

transfer_model.eval()
transfer_model = transfer_model.cuda()

batch_size = 16

ok_list = []
for length in range(1, 1000):
    x = torch.zeros(batch_size, 3, length, length, requires_grad=False).cuda()
    print('length = {}'.format(length))
    try:
        torch_out = transfer_model(x)
        print('length = {} OK!!!!'.format(length))
        ok_list.append(length)
        # break
    except:
        continue

print(ok_list)

hard test,When trying to enter the size,This network model can run through,Run through and test step by step.注意,Since there are some downsampling rounding operations in the neural network,So the size is in some range,For example, the output of the code just now is：

...
length = 998
length = 999
[193, 194, 195, 196, 197, 198, 199, 200, 201, 202, 203, 204, 205, 206, 207, 208, 209, 210, 211, 212, 213, 214, 215, 216, 217, 218, 219, 220, 221, 222, 223, 224]

尺寸224Just one of the acceptable sizes,It is also the original size of the designer.