Batch evaluation for multiple choice and open questions.

Evaluate given image using detected objects on the global metadata specifications. Assumptions: * Metadata combines 'include' clauses wit

Evaluate given image using detected objects on the global metadata specifications. Assumptions: * Metadata combines 'include' clauses wit

Evaluate a single image on specified metrics and return the results dict, including both scores and reasoning extracted from the GPT response

Evaluate images based on specified metrics: - Load annotation and images - Run GPT evaluation for each metric - Return scored results

Evaluate images for a specific model, category and image ID. Returns a dict with score and reasoning for each metric.

Evaluate one multi-element synthesis example (3 refs + 1 prediction). Returns a dict containing both score and reasoning for each metric.

Evaluate a single temporal‐prediction case: three reference frames + one predicted frame. Returns a dict mapping: metric -> score,

Send a single image to GPT for quality evaluation.

Call GPT for single-image evaluation (image_quality).

Extract consistency score and reasoning from GPT response.

Parse score and reason from JSON-formatted response

Exact all forms of numbers from a string with regex.

Extract quality score and reasoning from GPT response.

Use the Levenshtein distance (or edit distance) to determine which of the choices is most similar to the given prediction

This method allows to interpolate the pre-trained position encodings, to be able to use the model on higher resolution images. This m

Load an image from a given path or URL and convert it to a PIL Image. Args: image (Union[str, Image.Image]): The image path, URL, or