#4949th of 80 in Open models
sketch-code
CNN-LSTM model that turns hand-drawn wireframe images into HTML
- Stars
- 5.1k
- License
- MIT
- Last commit
- Aug 2026
- Language
- Python
5,124 on 22 Apr 20265,141 on 11 Nov 2026
From GitHub's daily star counts.
Checked on GitHub.
Overview
SketchCode is a 2018 proof-of-concept that treats HTML as the caption of a wireframe image, using a CNN encoder and LSTM decoder trained on a 1,700-image synthetic dataset derived from pix2code. The repo has Python scripts to convert single images or folders to HTML, train from scratch or from pretrained weights, and score output with BLEU. The README says the code targets TensorFlow 1.x and Keras and is no longer maintained.
Who it is for: Researchers and engineers studying early image-to-code models
Strengths
- Ships pretrained weights and a download script for the 1,700-image dataset (342 MB)
- CLI scripts cover single and batch conversion, training, and BLEU evaluation
- Training can start from scratch or from pretrained weights, with optional Keras augmentation
- MIT licensed, with a linked write-up of the approach
Weaknesses
- No longer maintained; targets TensorFlow 1.x and Keras of 2018
- Only works on wireframes resembling the synthetic training set
- No accuracy figures given; evaluation is a BLEU script only
- Portions derive from Apache-2.0 pix2code, adding attribution terms beyond MIT
What it needs
- Needs Python 3, TensorFlow 1.x, Keras
- Models: CNN encoder + LSTM decoder (Keras JSON + H5 weights)
Also in Open models
See all 80| Rank | Project | Score |
|---|---|---|
| 1 | ruviewWiFi CSI sensing on ESP32 for presence, breathing, heart rate and pose | 94 out of 100 |
| 2 | layaSingle-pass multilingual classifier returning typed decisions with calibrated odds | 88 out of 100 |
| 3 | rf-detrReal-time transformer models for object detection, segmentation and keypoints | 81 out of 100 |
| #4 | timesfmPretrained time-series foundation model for zero-shot forecasting from Google Research | 78 out of 100 |
| #5 | open_clipPyTorch library for training and running CLIP-style image-text models | 78 out of 100 |