AngleForge — Edge Impulse custom synthetic data block

Normalise the camera angle of the images already in your Edge Impulse project so the whole dataset is visually consistent (e.g. everything top-down). It runs the real Qwen-Image-Edit multi-angle model via the public AngleForge Hugging Face Space (ZeroGPU), so the block itself needs no GPU.

Enterprise only. Custom synthetic data blocks require an Edge Impulse Enterprise plan.

How it works

For every existing image sample it:

1. Lists your samples via the Studio API (GET /api/{projectId}/raw-data).
2. Downloads the image (GET /api/{projectId}/raw-data/{id}/image).
3. Sends it to the AngleForge Space grab_viewpoints API for the selected angle.
4. Uploads the re-rendered image back through the Ingestion API with the x-synthetic-data-job-id header (so it previews under Data acquisition → Synthetic data).

Original samples are left untouched; normalised copies are added with a generated_by=angleforge metadata tag and filename label.angle_<angle>.<id>.png.

Parameters

Param	Description
angle	Target camera angle preset (top_down, birds_eye, rotate_left_45, …).
angleforge-space	HF Space id running the model (default eoinedge/angleforge).
HF_TOKEN	Secret. Only needed if the Space is private.
image-size	Longest side of the output image (px).
source-category	Which images to normalise: training / testing / all.
upload-category	Where to put results: same / split / training / testing.
labels	Optional comma-separated label filter.
max-samples	Cap the number of images processed (0 = all).
seed	Reproducible generation seed.

Test locally

Synthetic data blocks are not supported by edge-impulse-blocks runner, so build and run the container directly:

docker build -t angleforge-synthetic .
docker run --rm \
  -e EI_PROJECT_ID='<project-id>' \
  -e EI_API_KEY='ei_...'             `# org or project API key (Studio reads)` \
  -e EI_PROJECT_API_KEY='ei_...'     `# project API key (ingestion writes)` \
  -e EI_INGESTION_HOST='edgeimpulse.com' \
  -e EI_API_ENDPOINT='https://studio.edgeimpulse.com/v1' \
  angleforge-synthetic \
  --synthetic-data-job-id 0 \
  --angle top_down --source-category training --upload-category training \
  --max-samples 3

Push to Edge Impulse

edge-impulse-blocks init      # choose: Synthetic data block
edge-impulse-blocks push

Then use it from Data acquisition → Synthetic data in any project in your organization.