Sanrio Classifier — ResNet18
ResNet18 fine-tuned to classify 29 Sanrio characters from images. Exported to ONNX for lightweight CPU inference.
Model Description
- Architecture: ResNet18 (Transfer Learning)
- Input: 224x224 RGB images
- Output: 29 Sanrio character classes with confidence scores
- Format: ONNX (FP32 and INT8 quantized)
- Size: 43 MB (FP32) / 11 MB (INT8)
Classes
badtz_maru, chococat, cinnamon, cinnamoroll, gudetama, hangyodon, hello_kitty, kerokerokeroppi, kuromi, little_twin_stars, melody, my_sweet_piano, pochacco, pompompurin, purin, rilakkuma, shin_chan, sumikko_gurashi, tuxedo_sam
Usage (Python)
`python import onnxruntime as ort import numpy as np from PIL import Image
sess = ort.InferenceSession('tl_sanrio_v5_int8.onnx') img = Image.open('character.jpg').resize((224, 224)) input_tensor = np.array(img).transpose(2,0,1).astype(np.float32) / 255.0 input_tensor = np.expand_dims(input_tensor, axis=0)
outputs = sess.run(['softmax_1'], {'input.1': input_tensor}) pred_class = outputs[0].argmax() confidence = outputs[0].max() `
Demo
Try it live: https://huggingface.co/spaces/MorritaConP1to/sanrio-classifier
Training
- Framework: PyTorch with Transfer Learning
- Optimizer: SGD with CosineAnnealing LR
- Dataset: ~6,160 images across 29 classes
- Hardware: NVIDIA RTX 4060