KiroSamurai/decore-vgg16-cifar10

DECORE — Compressed VGG-16 (CIFAR-10)

Channel-pruned VGG-16 produced with DECORE (arXiv:2106.06091): a reinforcement-learning method that assigns one agent per channel to learn which channels to drop. The dropped channels are physically removed.

Metric	Full VGG-16	DECORE-pruned	Reduction
Parameters	14.72 M	5.43 M	63.1%
FLOPs (MACs)	313.2 M	192.0 M	38.7%
Top-1 accuracy	93.48%	93.76%	+0.28%

No accuracy loss — VGG-16 is over-parameterized for CIFAR-10, so pruning acts as a regularizer.

Files

• pruned_vgg16_l50.pt — the compressed model (a pickled decore.models.VGGCifar).
• baseline.pth — the full VGG-16 baseline (state_dict) for comparison.

Usage

The pruned model is a standard PyTorch module but relies on the decore package for its class definition. Install the package (from the project repo), then:

import torch
pruned = torch.load("pruned_vgg16_l50.pt", map_location="cpu", weights_only=False).eval()

# CIFAR-10 preprocessing
import torchvision.transforms as T
tf = T.Compose([
    T.Resize((32, 32)), T.ToTensor(),
    T.Normalize([0.4914, 0.4822, 0.4465], [0.2470, 0.2435, 0.2616]),
])
logits = pruned(tf(img).unsqueeze(0))   # img: a PIL.Image

Classes: airplane, automobile, bird, cat, deer, dog, frog, horse, ship, truck.

Reproduction

See the project repository: python -m decore.cli run --config configs/vgg16_cifar10.yaml.

Citation

Alwani, Wang, Madhavan. DECORE: Deep Compression with Reinforcement Learning. CVPR 2022.