SpeciesNet Studio — self-hosted review UI for camera trap predictions

[arunrajiah]

Hi everyone,

I have built a free, open-source review interface for teams running camera trap AI (SpeciesNet, MegaDetector, or any model that writes a predictions JSON) and wanted to share it here.

The gap it fills: your model produces predictions — but a human reviewer still needs to confirm correct IDs, override wrong calls, and flag uncertain images before results go into a database or report. Doing that at scale in a spreadsheet or raw JSON is painfully slow.

What SpeciesNet Studio does:

• Points at a local folder, walks subdirectories, generates thumbnails, shows a virtualized gallery (smooth at 10 000+ images)
• Runs inference via the SpeciesNet Python API directly, or accepts output from any subprocess that writes SpeciesNet-compatible JSON — including MegaDetector pipelines
• Per-image review: Approve / Override (with species label + reviewer note) / Flag. Full keyboard shortcuts for speed
• Batch review — select a confidence sweep and approve or flag hundreds of frames in two clicks
• Live stats bar: confirmed / overridden / flagged / unreviewed / avg confidence per collection
• Export to CSV or JSON

Completely on-premise. No cloud account, no data upload, runs on your own hardware.

Two-command start:

git clone https://github.com/arunrajiah/speciesnet-studio
cd speciesnet-studio
docker compose -f docker-compose.release.yml up

Open http://localhost:8000 — sample wildlife images included, no setup needed to see it in action.

Already have predictions? If you have an existing predictions.json from MegaDetector or SpeciesNet, you can import it immediately using the subprocess adapter — no re-running the model.

Would love feedback from people actually processing camera trap data. What review workflows are missing? What export formats do your programs need?

👉 github.com/arunrajiah/speciesnet-studio

[ToltHill]

This sounds fantastic Arun! Looking forward to trying it. Rob Sent from Samsung Galaxy smartphone. Get Outlook for Android<https://aka.ms/AAb9ysg>

…

________________________________ From: Arun Rajiah ***@***.***> Sent: Wednesday, April 22, 2026 8:04:05 PM To: microsoft/CameraTraps ***@***.***> Cc: Subscribed ***@***.***> Subject: [microsoft/CameraTraps] SpeciesNet Studio — self-hosted review UI for camera trap predictions (Discussion #631) Hi everyone, I have built a free, open-source review interface for teams running camera trap AI (SpeciesNet, MegaDetector, or any model that writes a predictions JSON) and wanted to share it here. The gap it fills: your model produces predictions — but a human reviewer still needs to confirm correct IDs, override wrong calls, and flag uncertain images before results go into a database or report. Doing that at scale in a spreadsheet or raw JSON is painfully slow. What SpeciesNet Studio does: * Points at a local folder, walks subdirectories, generates thumbnails, shows a virtualized gallery (smooth at 10 000+ images) * Runs inference via the SpeciesNet Python API directly, or accepts output from any subprocess that writes SpeciesNet-compatible JSON — including MegaDetector pipelines * Per-image review: Approve / Override (with species label + reviewer note) / Flag. Full keyboard shortcuts for speed * Batch review — select a confidence sweep and approve or flag hundreds of frames in two clicks * Live stats bar: confirmed / overridden / flagged / unreviewed / avg confidence per collection * Export to CSV or JSON Completely on-premise. No cloud account, no data upload, runs on your own hardware. Two-command start: git clone https://github.com/arunrajiah/speciesnet-studio cd speciesnet-studio docker compose -f docker-compose.release.yml up Open http://localhost:8000 — sample wildlife images included, no setup needed to see it in action. Already have predictions? If you have an existing predictions.json from MegaDetector or SpeciesNet, you can import it immediately using the subprocess adapter — no re-running the model. Would love feedback from people actually processing camera trap data. What review workflows are missing? What export formats do your programs need? 👉 github.com/arunrajiah/speciesnet-studio<https://github.com/arunrajiah/speciesnet-studio> — Reply to this email directly, view it on GitHub<#631>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/ASRJAU342XNVXDS6PLHJREL4XGB2LAVCNFSM6AAAAACYDEBBI6VHI2DSMVQWIX3LMV43ERDJONRXK43TNFXW4OZZHEZTGNJZGY>. You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>