|
|
--- |
|
|
license: apache-2.0 |
|
|
datasets: |
|
|
- jonathan-roberts1/RSI-CB256 |
|
|
language: |
|
|
- en |
|
|
base_model: |
|
|
- google/siglip2-base-patch16-224 |
|
|
pipeline_tag: image-classification |
|
|
library_name: transformers |
|
|
tags: |
|
|
- Remote Sensing Instruments |
|
|
- RSI |
|
|
- Location |
|
|
--- |
|
|
|
|
|
 |
|
|
|
|
|
# **RSI-CB256-07** |
|
|
|
|
|
> **RSI-CB256-07** is a SigLIP2-based model fine-tuned for **coarse-grained remote sensing land-cover classification**. It distinguishes among 7 essential categories commonly used in environmental, urban planning, and geospatial analysis applications. The model is built on `google/siglip2-base-patch16-224 ` using the `SiglipForImageClassification` architecture. |
|
|
|
|
|
```py |
|
|
Classification Report: |
|
|
precision recall f1-score support |
|
|
|
|
|
transportation 0.9810 0.9858 0.9834 3300 |
|
|
other objects 0.9854 0.9932 0.9893 884 |
|
|
woodland 0.9973 0.9958 0.9966 6258 |
|
|
water area 0.9870 0.9837 0.9854 4104 |
|
|
other land 0.9925 0.9919 0.9922 3593 |
|
|
cultivated land 0.9918 0.9901 0.9909 2817 |
|
|
construction land 0.9945 0.9963 0.9954 3791 |
|
|
|
|
|
accuracy 0.9912 24747 |
|
|
macro avg 0.9899 0.9910 0.9904 24747 |
|
|
weighted avg 0.9912 0.9912 0.9912 24747 |
|
|
``` |
|
|
|
|
|
 |
|
|
|
|
|
--- |
|
|
|
|
|
## **Label Space: 7 Remote Sensing Classes** |
|
|
|
|
|
This model predicts one of the following categories for a given satellite or aerial image: |
|
|
|
|
|
``` |
|
|
Class 0: "transportation" |
|
|
Class 1: "other objects" |
|
|
Class 2: "woodland" |
|
|
Class 3: "water area" |
|
|
Class 4: "other land" |
|
|
Class 5: "cultivated land" |
|
|
Class 6: "construction land" |
|
|
``` |
|
|
|
|
|
--- |
|
|
|
|
|
## **Install Dependencies** |
|
|
|
|
|
```bash |
|
|
pip install -q transformers torch pillow gradio |
|
|
``` |
|
|
|
|
|
--- |
|
|
|
|
|
## **Inference Code** |
|
|
|
|
|
```python |
|
|
import gradio as gr |
|
|
from transformers import AutoImageProcessor, SiglipForImageClassification |
|
|
from PIL import Image |
|
|
import torch |
|
|
|
|
|
# Load model and processor |
|
|
model_name = "prithivMLmods/RSI-CB256-07" |
|
|
model = SiglipForImageClassification.from_pretrained(model_name) |
|
|
processor = AutoImageProcessor.from_pretrained(model_name) |
|
|
|
|
|
# ID to label mapping |
|
|
id2label = { |
|
|
"0": "transportation", |
|
|
"1": "other objects", |
|
|
"2": "woodland", |
|
|
"3": "water area", |
|
|
"4": "other land", |
|
|
"5": "cultivated land", |
|
|
"6": "construction land" |
|
|
} |
|
|
|
|
|
def classify_rsi_image(image): |
|
|
image = Image.fromarray(image).convert("RGB") |
|
|
inputs = processor(images=image, return_tensors="pt") |
|
|
|
|
|
with torch.no_grad(): |
|
|
outputs = model(**inputs) |
|
|
logits = outputs.logits |
|
|
probs = torch.nn.functional.softmax(logits, dim=1).squeeze().tolist() |
|
|
|
|
|
prediction = { |
|
|
id2label[str(i)]: round(probs[i], 3) for i in range(len(probs)) |
|
|
} |
|
|
|
|
|
return prediction |
|
|
|
|
|
# Gradio Interface |
|
|
iface = gr.Interface( |
|
|
fn=classify_rsi_image, |
|
|
inputs=gr.Image(type="numpy"), |
|
|
outputs=gr.Label(num_top_classes=7, label="Predicted Land-Cover Category"), |
|
|
title="RSI-CB256-07", |
|
|
description="Upload a satellite or aerial image to classify it into one of seven coarse land-cover classes using SigLIP2." |
|
|
) |
|
|
|
|
|
if __name__ == "__main__": |
|
|
iface.launch() |
|
|
``` |
|
|
|
|
|
--- |
|
|
|
|
|
## **Applications** |
|
|
|
|
|
* **Urban vs Rural Segmentation** |
|
|
* **Land-Use Classification** |
|
|
* **National/Regional Land Cover Monitoring** |
|
|
* **Environmental Impact Assessment** |
