TrapTracker Offline — Batch Camera-Trap Image & Video Processing
Process your entire camera-trap archive in one pass. TrapTracker Offline is a Docker-based web application that runs AI species detection across large batches of images and videos on your own machine, and gives you a clean, structured record of everything it finds — no cloud, no uploads, no per-image fees.
It’s built for conservationists, ecologists and biodiversity practitioners who come back from the field with SD cards full of footage and need to turn thousands — or hundreds of thousands — of files into usable data.
What it does
Reviewing camera-trap footage by hand is slow, and it’s where most projects lose time. TrapTracker Offline closes that gap. You point it at a folder of images and videos, choose your model, and the app:
- Runs each image and video frame through an ONNX AI detection model
- Records every detection — species, confidence and location in the frame — to a CSV
- Saves annotated copies of the images so you can verify results at a glance
- Extracts the original EXIF capture date from each photo into the CSV, so detections stay tied to when they were taken
Everything runs through a simple browser-based interface: add your files, pick a model and its class list, set your confidence threshold, and start the run. Progress updates live as it works.
Key features
- Built for very large batches — comfortably handles hundreds of thousands of images and videos in a single run, writing results to disk as it goes rather than holding everything in memory
- Automatic resume — if a run is interrupted by a crash, restart or power loss, pick up exactly where it left off; already-processed files are skipped and nothing already written is lost
- Images and video — processes common image formats (JPEG, PNG, TIFF, BMP, WebP) and video (MP4, MOV, AVI, MKV), with per-frame detection for clips
- Choose your own models — drop in the detection model and class list for your region or study; the app discovers the models you provide and lets you pick per run
- Structured CSV output — one row per detection with species, confidence, bounding box, source file and capture date, ready for Excel, R or QGIS
- Annotated image gallery — review labelled results in the built-in gallery, or download everything as a single ZIP
- GPU acceleration — optional NVIDIA GPU support for high throughput, with a clear warning if the app falls back to CPU so you’re never unknowingly running slow
- Your data stays yours — everything runs locally in Docker; nothing is sent anywhere
How it works
Copy your images and videos into the input folder, open the web interface, and select the ONNX model and matching class-names file for your project. Set a confidence threshold, choose whether to save annotated images, and start the run. As each file is processed, its detections are streamed to the results CSV and logged, so the record on disk is always up to date. When the run finishes you get a species summary, a downloadable CSV, annotated images in the gallery, and a ZIP of the full results. A separate CSV lists any files that couldn’t be read, so nothing fails silently.
Runs on your own hardware
TrapTracker Offline is deployed with Docker and runs entirely on your own machine or server. It works with an NVIDIA GPU for acceleration (built against CUDA 12.4 / cuDNN 9); on a machine without a compatible GPU it will run on CPU. For privacy and security, the application runs locally and is not designed to be exposed to the public internet.
Free for conservation and research
TrapTracker Offline is free to use for non-commercial conservation, ecological research, biodiversity monitoring, academic research and education.
Commercial use — paid consultancy, resale, hosted services, or integration into proprietary commercial platforms — requires prior written permission.
Get it on GitHub
The full source, Docker files and step-by-step tutorial are available on GitHub:
https://github.com/traptracker-ai/traptracker-offline-processor
AI model weights are released separately through Zenodo so they can be cited with a DOI. Please select the appropriate model from the Downloads/Models directory for specific model information and download link.