Practical overview
When it fits
Use it as a reference or prototype when the target classes are stop signs and pedestrian crossing signs and a TensorFlow SSD training and inference workflow is relevant.
Who it is for
- Developers studying SSD-based object detection in TensorFlow.
- Researchers or prototypers exploring traffic-sign detection with the LISA dataset.
Documented uses
- Prepare a selected subset of LISA traffic-sign data for the training pipeline.
- Train an SSD model for stop and pedestrian crossing sign detection.
- Run inference on sample images or provide image and video inputs to the inference script.
Requirements & limits
The repository is public and uses the MIT license.
- The model currently detects only stop signs and pedestrian crossing signs.
- The project reports overfitting and describes itself as a work in progress.
- The README reports 40–45 fps on a GTX 1080 with an Intel Core i7-6700K; in that setup, CPU-only non-maximum suppression took longer than neural-network inference.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Prepare data, train the detector and run sample inference
Use the documented LISA dataset workflow to train the model and display annotated sample images.
Before you start
- Python 3.5 or later
- TensorFlow 0.12.0, Pickle and OpenCV-Python; Matplotlib is optional
- The LISA Traffic Sign Dataset and a local clone of the repository
Steps
Prepare the LISA dataset's mergedAnnotations.csv to include only stop signs and pedestrian crossing signs.
Copy data_gathering/create_pickle.py into the dataset directory and run it there.
From the repository, link the generated resized image and raw pickle files, then run data_prep.py.
Run train.py, followed by inference.py -m demo to annotate and display images from sample_images.
Expected outcome
The data is packaged for the training pipeline, the SSD model is trained, and the demo displays annotated sample images.
Guide limitations
- The project reports severe overfitting.
- The model currently detects only stop signs and pedestrian crossing signs.
Run inference on supplied images or videos
Pass image or video input to the inference script for traffic-sign detection.
Before you start
- The repository's inference.py script and a trained model
- An image or video input
Steps
Use the inference.py script's -i flag to specify your own image or video input.
Consult the script's code for the flag's detailed usage.
Expected outcome
The inference script processes the supplied input for traffic-sign predictions.
Guide limitations
- The README warns that the model severely overfits.
- Only stop signs and pedestrian crossing signs are currently detected.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 17 October 2026
- Original website title
- georgesung/ssd_tensorflow_traffic_sign_detection · Official homepage title · retained in its source language