Bug: ResNet18 final layer hardcoded to 1000 classes instead of using num_classes argument

[abhiram123467]

Bug Description

In the ResNet18 model definition, num_classes is accepted
as an argument in __init__ but is completely ignored.
The final fully connected layer is hardcoded to output
1000 classes (ImageNet default), which causes silent
wrong behavior when the model is used for Renaissance OCR
classification tasks.

Location

File: [paste exact file path here]

Current Broken Code

class ResNet18(nn.Module):
    def __init__(self, num_classes):  # accepted but ignored!
        super().__init__()
        self.fc = nn.Linear(512, 1000)  # hardcoded 1000 — BUG!

Fixed Code

class ResNet18(nn.Module):
    def __init__(self, num_classes):
        super().__init__()
        self.fc = nn.Linear(512, num_classes)  # uses argument correctly

Impact

• Model silently outputs wrong number of classes
• Will crash or give wrong results for any dataset
  with classes ≠ 1000
• Renaissance OCR tasks have far fewer than 1000 classes

Proposed Fix

Replace hardcoded 1000 with num_classes in the
Linear layer definition.

I would like to submit a PR to fix this.

[aligul-dev]

Hi, I looked into this issue and it seems the ResNet18 final layer is hardcoded to 1000 classes.
I believe it should instead use the num_classes parameter so the model can work with different datasets.
I would like to work on fixing this if no one else is currently assigned.

[agentksimha]

class ResNet18(nn.Module):
def init(self, num_classes): # accepted but ignored!
super().init()
self.fc = nn.Linear(512, 1000) # hardcoded 1000 — BUG! where is this code snippet ? IN the Resnet model i refactored(which i made a PR for recently) i didnt observe this code block .

[alimohmedelsaid26-cell]

Hi, I've reviewed the hardcoded num_classes in the ResNet18 implementation. I would like to fix this by making the final fully connected layer dynamic based on the num_classes parameter. I’ve already forked the repo and I'm ready to submit a PR. Can you assign this to me?