Spaces:
Running
on
Zero
Running
on
Zero
Ready to leverage your Real-ESRGAN-Anime-finetuning model for breathtaking anime image upscaling based on Real-ESRGAN_x4Plus?
#17
by
danhtran2mind
- opened
Overview π
Real-ESRGAN-Anime-finetuning, built upon Real-ESRGAN_x4Plus, is optimized for high-quality anime image upscaling, delivering superior clarity and consistency! π
Performance Comparison π
Model Name | Meanβ | Std Deviationβ |
---|---|---|
RealESRGAN_x4plus | 36.8437 | 3.4551 |
Real-ESRGAN-Anime-finetuning | 36.8695 π | 3.3193 π |
PSNR Results π
Input File | Upscale Factor | Real-ESRGAN-Anime-finetuning | RealESRGAN_x4plus_anime_6B |
---|---|---|---|
_resized_div_2.png | 2 | 27.65 dB π | 26.01 dB |
_resized_div_4.png | 4 | 24.46 dB π | 23.07 dB |
Requirements π§
This model is based on Real-ESRGAN_x4Plus and requires its architecture for optimal performance. Ensure you have the necessary dependencies for Real-ESRGAN_x4Plus. π οΈ
Usage πΌοΈ
1. Download Model π₯
Get the model from:
https://huggingface.co/danhtran2mind/Real-ESRGAN-Anime-finetuning/resolve/main/Real-ESRGAN-Anime-finetuning.pth
2. Install stablepy
π¦
pip install stablepy
3. Run Inference π»
from PIL import Image
from stablepy import load_upscaler_model
scaler = load_upscaler_model(
model="./Real-ESRGAN-Anime-finetuning.pth",
tile=0,
tile_overlap=8,
device="cuda",
half=True
)
image = Image.open("<your_image_path>").convert("RGB")
upscaled = scaler.upscale(image, 4)
upscaled
Why This Model? π€©
- Based on Real-ESRGAN_x4Plus: Leverages a proven architecture for anime optimization. ποΈ
- Top Metrics: Higher quality and consistency. π
- Easy Integration: Seamless use with
stablepy
. π
Upscale your anime images with this enhanced model! πβ¨