How 58.com Built an AI Designer: From Smart Cutout to Intelligent Creative Platform

This article chronicles 58.com’s journey from a small brainstorming room to a full‑scale AI design platform, detailing the development of smart cutout, the BASNet segmentation model, custom loss functions, template editing, and the measurable business impact of the AI designer.

58UXD
58UXD
58UXD
How 58.com Built an AI Designer: From Smart Cutout to Intelligent Creative Platform

Origin of the AI Designer

In December 2019, a team at 58.com’s User Experience Lab gathered in a 20‑square‑meter room to discuss creating an AI designer that could serve the company’s business needs.

First version during the pandemic

In February 2020 the smart cutout capability was launched, reducing processing time from 5 minutes per 10 images to 5 seconds and eventually handling up to 172,800 images per day.

To address the heavy demand of the used‑car business, the team built a backend pipeline that cut manual effort dramatically, saving roughly 1.5 million RMB in labor costs.

Technical foundation – BASNet

The AI designer’s image segmentation relies on the Boundary‑Aware Salient Object Detection (BASNet) model, chosen for its ability to preserve object edges and completeness. The architecture consists of an encoder‑decoder for coarse segmentation and a residual refinement module for fine adjustments.

Two custom loss functions are employed: Ranking Loss, which pushes the network to enlarge score gaps when predictions are close, and Saliency‑Aware Loss, which penalizes mask errors in non‑salient regions while allowing flexibility in salient areas.

“Smart cutout belongs to deep learning’s salient object detection (SOD) field and is a hot research area.” – Wei Lei

Template editor and intelligent generation

The front‑end editor parses design files to extract bounding‑box coordinates, while the back‑end reconstructs visual effects. Integrated with smart cutout, the platform adds features such as background blur for used‑car listings and automatic shadow generation for new‑car images.

Combined with the intelligent generation module, the system can produce up to 14,400 new creative assets per day, distinguishing foreground from background and placing new elements without occluding the main subject.

Business impact

Since its launch, the AI designer has processed nearly 300,000 vehicle images, saving about 150 万元 in labor costs and improving design efficiency by up to 43 % across quarters.

“The smart cutout service saves about 3,000 images per day for the used‑car business, protecting information security while improving reviewer experience.” – Automotive Technology Dept
Original Source

Signed-in readers can open the original source through BestHub's protected redirect.

Sign in to view source
Republication Notice

This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactadmin@besthub.devand we will review it promptly.

Computer VisionDeep Learningimage segmentationAI designBASNetsmart cutouttemplate editor
58UXD
Written by

58UXD

58.com User Experience Design Center

0 followers
Reader feedback

How this landed with the community

Sign in to like

Rate this article

Was this worth your time?

Sign in to rate
Discussion

0 Comments

Thoughtful readers leave field notes, pushback, and hard-won operational detail here.