How to load resnet in new fastai version

Part 1 2022
1

I am probably doing a very silly mistake. i can’t load my trained model . Here is the error:


FileNotFoundError: [Errno 2] No such file or directory: 'data/models/<function resnet18 at 0x7f2ca2cbdb70>.pth'

why is it a function. I have imported is as well
I am using the ensemble notebook that is on github

Thanks in advance

2

Hey, you probably just forgot quotes around resnet18. To receive more specific help you would need to describe how you got to that error.

3

having (resnet18) in quotes will throw this error:

RuntimeError: Error(s) in loading state_dict for Sequential:
	Missing key(s) in state_dict: "0.model.conv1.weight", "0.model.bn1.weight", "0.model.bn1.bias", "0.model.bn1.running_mean", "0.model.bn1.running_var", "0.model.layer1.0.conv1.weight", "0.model.layer1.0.bn1.weight", "0.model.layer1.0.bn1.bias", "0.model.layer1.0.bn1.running_mean", "0.model.layer1.0.bn1.running_var", "0.model.layer1.0.conv2.weight", "0.model.layer1.0.bn2.weight", "0.model.layer1.0.bn2.bias", "0.model.layer1.0.bn2.running_mean", "0.model.layer1.0.bn2.running_var", "0.model.layer1.1.conv1.weight", "0.model.layer1.1.bn1.weight", "0.model.layer1.1.bn1.bias", "0.model.layer1.1.bn1.running_mean", "0.model.layer1.1.bn1.running_var", "0.model.layer1.1.conv2.weight", "0.model.layer1.1.bn2.weight", "0.model.layer1.1.bn2.bias", "0.model.layer1.1.bn2.running_mean", "0.model.layer1.1.bn2.running_var", "0.model.layer2.0.conv1.weight", "0.model.layer2.0.bn1.weight", "0.model.layer2.0.bn1.bias", "0.model.layer2.0.bn1.running_mean", "0.model.layer2.0.bn1.running_var", "0.model.layer2.0.conv2.weight", "0.model.layer2.0.bn2.weight", "0.model.layer2.0.bn2.bias", "0.model.layer2.0.bn2.running_mean", "0.model.layer2.0.bn2.running_var", "0.model.layer2.0.downsample.0.weight", "0.model.layer2.0.downsample.1.weight", "0.model.layer2.0.downsample.1.bias", "0.model.layer2.0.downsample.1.running_mean", "0.model.layer2.0.downsample.1.running_var", "0.model.layer2.1.conv1.weight", "0.model.layer2.1.bn1.weight", "0.model.layer2.1.bn1.bias", "0.model.layer2.1.bn1.running_mean", "0.model.layer2.1.bn1.running_var", "0.model.layer2.1.conv2.weight", "0.model.layer2.1.bn2.weight", "0.model.layer2.1.bn2.bias", "0.model.layer2.1.bn2.running_mean", "0.model.layer2.1.bn2.running_var", "0.model.layer3.0.conv1.weight", "0.model.layer3.0.bn1.weight", "0.model.layer3.0.bn1.bias", "0.model.layer3.0.bn1.running_mean", "0.model.layer3.0.bn1.running_var", "0.model.layer3.0.conv2.weight", "0.model.layer3.0.bn2.weight", "0.model.layer3.0.bn2.bias", "0.model.layer3.0.bn2.running_mean", "0.model.layer3.0.bn2.running_var", "0.model.layer3.0.downsample.0.weight", "0.model.layer3.0.downsample.1.weight", "0.model.layer3.0.downsample.1.bias", "0.model.layer3.0.downsample.1.running_mean", "0.model.layer3.0.downsample.1.running_var", "0.model.layer3.1.conv1.weight", "0.model.layer3.1.bn1.weight", "0.model.layer3.1.bn1.bias", "0.model.layer3.1.bn1.running_mean", "0.model.layer3.1.bn1.running_var", "0.model.layer3.1.conv2.weight", "0.model.layer3.1.bn2.weight", "0.model.layer3.1.bn2.bias", "0.model.layer3.1.bn2.running_mean", "0.model.layer3.1.bn2.running_var", "0.model.layer4.0.conv1.weight", "0.model.layer4.0.bn1.weight", "0.model.layer4.0.bn1.bias", "0.model.layer4.0.bn1.running_mean", "0.model.layer4.0.bn1.running_var", "0.model.layer4.0.conv2.weight", "0.model.layer4.0.bn2.weight", "0.model.layer4.0.bn2.bias", "0.model.layer4.0.bn2.running_mean", "0.model.layer4.0.bn2.running_var", "0.model.layer4.0.downsample.0.weight", "0.model.layer4.0.downsample.1.weight", "0.model.layer4.0.downsample.1.bias", "0.model.layer4.0.downsample.1.running_mean", "0.model.layer4.0.downsample.1.running_var", "0.model.layer4.1.conv1.weight", "0.model.layer4.1.bn1.weight", "0.model.layer4.1.bn1.bias", "0.model.layer4.1.bn1.running_mean", "0.model.layer4.1.bn1.running_var", "0.model.layer4.1.conv2.weight", "0.model.layer4.1.bn2.weight", "0.model.layer4.1.bn2.bias", "0.model.layer4.1.bn2.running_mean", "0.model.layer4.1.bn2.running_var". 
	Unexpected key(s) in state_dict: "0.0.weight", "0.1.weight", "0.1.bias", "0.1.running_mean", "0.1.running_var", "0.1.num_batches_tracked", "0.4.0.conv1.weight", "0.4.0.bn1.weight", "0.4.0.bn1.bias", "0.4.0.bn1.running_mean", "0.4.0.bn1.running_var", "0.4.0.bn1.num_batches_tracked", "0.4.0.conv2.weight", "0.4.0.bn2.weight", "0.4.0.bn2.bias", "0.4.0.bn2.running_mean", "0.4.0.bn2.running_var", "0.4.0.bn2.num_batches_tracked", "0.4.1.conv1.weight", "0.4.1.bn1.weight", "0.4.1.bn1.bias", "0.4.1.bn1.running_mean", "0.4.1.bn1.running_var", "0.4.1.bn1.num_batches_tracked", "0.4.1.conv2.weight", "0.4.1.bn2.weight", "0.4.1.bn2.bias", "0.4.1.bn2.running_mean", "0.4.1.bn2.running_var", "0.4.1.bn2.num_batches_tracked", "0.5.0.conv1.weight", "0.5.0.bn1.weight", "0.5.0.bn1.bias", "0.5.0.bn1.running_mean", "0.5.0.bn1.running_var", "0.5.0.bn1.num_batches_tracked", "0.5.0.conv2.weight", "0.5.0.bn2.weight", "0.5.0.bn2.bias", "0.5.0.bn2.running_mean", "0.5.0.bn2.running_var", "0.5.0.bn2.num_batches_tracked", "0.5.0.downsample.0.weight", "0.5.0.downsample.1.weight", "0.5.0.downsample.1.bias", "0.5.0.downsample.1.running_mean", "0.5.0.downsample.1.running_var", "0.5.0.downsample.1.num_batches_tracked", "0.5.1.conv1.weight", "0.5.1.bn1.weight", "0.5.1.bn1.bias", "0.5.1.bn1.running_mean", "0.5.1.bn1.running_var", "0.5.1.bn1.num_batches_tracked", "0.5.1.conv2.weight", "0.5.1.bn2.weight", "0.5.1.bn2.bias", "0.5.1.bn2.running_mean", "0.5.1.bn2.running_var", "0.5.1.bn2.num_batches_tracked", "0.6.0.conv1.weight", "0.6.0.bn1.weight", "0.6.0.bn1.bias", "0.6.0.bn1.running_mean", "0.6.0.bn1.running_var", "0.6.0.bn1.num_batches_tracked", "0.6.0.conv2.weight", "0.6.0.bn2.weight", "0.6.0.bn2.bias", "0.6.0.bn2.running_mean", "0.6.0.bn2.running_var", "0.6.0.bn2.num_batches_tracked", "0.6.0.downsample.0.weight", "0.6.0.downsample.1.weight", "0.6.0.downsample.1.bias", "0.6.0.downsample.1.running_mean", "0.6.0.downsample.1.running_var", "0.6.0.downsample.1.num_batches_tracked", "0.6.1.conv1.weight", "0.6.1.bn1.weight", "0.6.1.bn1.bias", "0.6.1.bn1.running_mean", "0.6.1.bn1.running_var", "0.6.1.bn1.num_batches_tracked", "0.6.1.conv2.weight", "0.6.1.bn2.weight", "0.6.1.bn2.bias", "0.6.1.bn2.running_mean", "0.6.1.bn2.running_var", "0.6.1.bn2.num_batches_tracked", "0.7.0.conv1.weight", "0.7.0.bn1.weight", "0.7.0.bn1.bias", "0.7.0.bn1.running_mean", "0.7.0.bn1.running_var", "0.7.0.bn1.num_batches_tracked", "0.7.0.conv2.weight", "0.7.0.bn2.weight", "0.7.0.bn2.bias", "0.7.0.bn2.running_mean", "0.7.0.bn2.running_var", "0.7.0.bn2.num_batches_tracked", "0.7.0.downsample.0.weight", "0.7.0.downsample.1.weight", "0.7.0.downsample.1.bias", "0.7.0.downsample.1.running_mean", "0.7.0.downsample.1.running_var", "0.7.0.downsample.1.num_batches_tracked", "0.7.1.conv1.weight", "0.7.1.bn1.weight", "0.7.1.bn1.bias", "0.7.1.bn1.running_mean", "0.7.1.bn1.running_var", "0.7.1.bn1.num_batches_tracked", "0.7.1.conv2.weight", "0.7.1.bn2.weight", "0.7.1.bn2.bias", "0.7.1.bn2.running_mean", "0.7.1.bn2.running_var", "0.7.1.bn2.num_batches_tracked".

Here is part of my code:

def load(arch, size, item=Resize(480, method='squish'), accum=1):
    dls = ImageDataLoaders.from_folder(trn_path,valid_pct=0.2, item_tfms=item,
        batch_tfms=aug_transforms(size=size, min_scale=0.75), bs=100//accum)
    gradient_accumulation = GradientAccumulation(64) if accum else []
    learn = vision_learner(dls, arch, metrics=error_rate, 
    cbs=[gradient_accumulation])
    learn.load(f'{arch}',with_opt=True)
    return learn.tta(dl=dls.test_dl(tst_files))

models = {
    
     resnet18: {
        (Resize(640, method='squish'), 480),
    }}

tta_res = []

for arch,details in models.items():
    for item,size in details:
        print('---',arch)
        print(type(arch))
        print(size)
        print(item.name)
        tta_res.append(load(arch, size, item=item, accum=2)) #, epochs=1))

Thanks for your help @benkarr

4

resnet18 is a function, when you pass this to learn.load(f'{arch}', …) it tries to load your model from <function resnet18 at 0x7f2ca2cbdb70> (thats what you get if you run print(resnet18)) where you probably didn’t save it. When you use quotes around resnet18: vision_learner(…) creates a ‘timm’ model (instead of a fastai one) and the parameters you saved probably do not fit timms resnet18 schema…

To solve this you could add something to your models dict, like:

models = {
    resnet18: {
        path: 'path/to/model',
        tfms: (Resize(640, method='squish'), 480),
    }
}

and adjust your load function to load from models[arch].path.
Also: how did you save your trained model? If you used learn.export(path) you can load it with learn = load_learner(path) and don’t even need the specific architecture & transforms to recreate it :slight_smile: .

4 Likes
5

Thanks a lot for your help!

I saved both the pkl and pth with learn.save and learn.export. I ended up loading the model this way: learn.load(model_path + '/' + arch.__name__) .
Your explanation was very informative!

Cheers!

1 Like