CS5242 Intel Image Classification — Model Checkpoints
Trained checkpoints for the NUS CS5242 group project (Group 13). Models classify 6-class natural scene images (buildings, forest, glacier, mountain, sea, street) from the Intel Image Classification dataset.
Files
| File | Architecture | Notes |
|---|---|---|
mlp_best.pt |
MLP baseline | Flattened-pixel input |
cnn_best.pt |
CNN baseline | Reference CNN |
cnn_2block_best.pt |
CNN, 2 conv blocks | Capacity ablation |
cnn_3block_best.pt |
CNN, 3 conv blocks | Capacity ablation |
cnn_dropout03_best.pt |
CNN + dropout 0.3 | Regularization ablation |
cnn_dropout07_best.pt |
CNN + dropout 0.7 | Regularization ablation |
cnn_strong_aug_best.pt |
CNN + strong augmentation | Augmentation ablation |
vit_best.pt |
Vision Transformer | Best-performing model |
*_history.json |
— | Per-epoch loss/accuracy curves |
Loading
import torch
from huggingface_hub import hf_hub_download
ckpt_path = hf_hub_download(repo_id="daryl336/cs5242-intel-classification",
filename="vit_best.pt")
state = torch.load(ckpt_path, map_location="cpu")