SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2203.13452 · CVPR · 2022

PCA-Based Knowledge Distillation Towards Lightweight and Content-Style Balanced Photorealistic Style Transfer Models

Danna Gurari, Tai-Yin Chiu

arXiv · PDF · Open in the Atlas

Code that ran

We have not lifted any functions out of this paper's repositories yet, so there is nothing we have run. The repositories linked to it are listed below.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Figure 1. Exemplification of our PCA-based knowledge distillation resulting in models that are more lightweight, faster, and achieve a better content-style balance for photorealistic style transfer than existing models. We demonstrate its versatility by reporting results for when our approach is applied to two backbone architectures: VGG (Ours-VGG) and MobileNet (Ours-Mob). Compared to WCT 2 , our smallest model uses only 0.7% of the parameters (73K vs. 10.12M) while producing results over 6x faster (0.04s vs. 0.27s). Additionally, our distilled models consistently transfer stronger style effects than WCT 2 [48] and preserve content better than PhotoWCT-HFR (an improved variant of PhotoWCT [24]), PhotoWCT 2 [11], and a CKD [43] distilled model.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2203.13452")
get_code_for_paper("2203.13452")
have("2203.13452")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2203.13452.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2203.13452.svg)](https://syntology.ai/paper/2203.13452)

Connect an agent — have() is free.