cellmap_flow.ai_annotate.images

Checking the images a model sends back, and encoding images for the browser.

A model’s response is untrusted input: its image could be huge, malformed or in a format whose decoder has a history of bugs. decode_model_image reads only the header first (PIL’s open is lazy), checks the format and size from it, and decodes the pixels only when both pass. Nothing the model returned goes to the browser as is: png_base64 re-encodes the decoded pixels as a fresh PNG.

Attributes

`MAX_PIXELS`
`ALLOWED_FORMATS`

Functions

`decode_model_image`(data)	The model's image as RGB, after checking its format and size before decoding.
`png_base64`(image)	`image` as base64 PNG text (no `data:` prefix), for a JSON response.

Module Contents

cellmap_flow.ai_annotate.images.MAX_PIXELS = 16777216

cellmap_flow.ai_annotate.images.ALLOWED_FORMATS = ('PNG', 'JPEG', 'WEBP')

cellmap_flow.ai_annotate.images.decode_model_image(data)

The model’s image as RGB, after checking its format and size before decoding.

Raises AIAnnotateError("bad_response") for anything other than a PNG, JPEG or WebP of at most MAX_PIXELS pixels, or one that fails to decode.

cellmap_flow.ai_annotate.images.png_base64(image): image as base64 PNG text (no data: prefix), for a JSON response.