Generic TFLite Object Detection Harness

Overview

This tool provides a clean, general-purpose Python runner for TensorFlow Lite (TFLite) object detection models.
It can execute inference over live camera feeds, video files, image directories, or session-style datasets, producing structured JSON reports and Markdown summaries.

All vendor-specific content and model references have been removed.
The harness is model-agnostic and can be used with any .tflite model and associated metadata you legally own or are licensed to use.

---

Features

• Supports YOLO, SSD, or compatible TFLite model architectures.
• Works with webcam streams, video files, image directories, or recursive session folders.
• Produces JSON detection reports per frame or per session.
• Generates an ongoing Markdown summary of inference statistics.
• Optional OpenCV visualization of real-time detections.
• Session watcher mode for continuous processing of incoming datasets.

---

Installation

1. Clone and prepare environment

git clone https://github.com/GainSec/tensorflow-generic-harness
cd generic-tflite-harness
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

### Alternative:
1. Create/activate a virtualenv and install prerequisites (run `./setup.sh`, which defaults to `/opt/homebrew/bin/python3.11` on macOS and `python3` on Linux; override with `PYTHON_BIN=/path/to/python ./setup.sh` if needed):
```bash
./setup.sh        # creates .venv + installs platform-specific deps
source .venv/bin/activate

How to Use

2. Example webcam run:

   python3 generic_tflite_harness.py --models-json assets/models/objects/models.json --assets-root assets/models/objects --model-name model_a_float16 --camera 0

3. Example session replay:

   python3 generic_tflite_harness.py --models-json assets/models/objects/models.json --assets-root assets/models/objects --model-name model_a_float16 --session-root /path/to/session/root --session-stage processed

CLI Flags

• --models-json PATH – path to a lawful, user-supplied models.json.
• --assets-root PATH – directory containing compatible .tflite or equivalent model files.
• --model-name NAME – model identifier from models.json.
• --camera INDEX, --video PATH, --frames-dir DIR, --session-root DIR, --session-stage NAME, and related flags behave as described above.

Disclaimer

Use only models that you own or are explicitly licensed to use. Do not distribute proprietary or third-party assets. This harness is provided strictly for research, education, and lawful interoperability testing.