---
language: en
license: mit
tags:
- image-classification
- resnet
- corrosion
library_name: pytorch
pipeline_tag: image-classification
task_categories:
- image-classification
dataset: custom
---

# Corrosion Classifier (ResNet50)

This repository contains a ResNet50 image classifier trained to detect corrosion types.

## Labels
- crevice_corrosion
- erosion_corrosion
- galvanic_corrosion
- mic_corrosion
- no_corrosion
- pitting_corrosion
- stress_corrosion
- under_insulation_corrosion
- uniform_corrosion

## Usage (PyTorch)

```python
import torch, json
from PIL import Image
from torchvision import transforms
import timm

# Load labels
labels = ['crevice_corrosion', 'erosion_corrosion', 'galvanic_corrosion', 'mic_corrosion', 'no_corrosion', 'pitting_corrosion', 'stress_corrosion', 'under_insulation_corrosion', 'uniform_corrosion']

# Create model
model = timm.create_model('resnet50', pretrained=False, num_classes=len(labels))
state = torch.load('resnet50-corrosion-classifier-v1.pth', map_location='cpu')
missing, unexpected = model.load_state_dict(state, strict=False)
model.eval()

# Preprocess (ImageNet)
transform = transforms.Compose([
    transforms.Resize(256, interpolation=transforms.InterpolationMode.BICUBIC),
    transforms.CenterCrop(224),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
])

img = Image.open('test.jpg').convert('RGB')
x = transform(img).unsqueeze(0)

with torch.no_grad():
    logits = model(x)
    probs = logits.softmax(dim=1).squeeze().tolist()

idx = int(torch.tensor(probs).argmax())
print(labels[idx], probs[idx])
```

> Note: This is a **generic PyTorch checkpoint** (`.pth`). The public Inference API on the Hub does **not** execute arbitrary PyTorch code. If you want to call this model via the **Inference API**, you must convert it to a supported library format (e.g. `transformers` image-classification) or use your existing **Space** and call it via the Gradio API. See below.

## Call via your existing Space (recommended now)
If your Space works, you can call it programmatically using the Gradio JS Client from Node:
```js
import { Client, handle_file } from "@gradio/client";

const app = await Client.connect("jacopo22295/RESNET50-CORROSION_CLASSIFIER_V1"); // your Space id
const res = await fetch("https://example.com/image.jpg");
const blob = await res.blob();
const out = await app.predict("/predict", [handle_file(blob)]);
console.log(out.data);
```

## Convert to Transformers (optional, to use Inference API)
If you later want to enable the one-click Inference API, consider exporting to a `transformers` ImageClassification model (e.g. `ResNetForImageClassification`) and pushing weights + `preprocessor_config.json`. This requires a small conversion script.