vision_unlearning.datasets.testbed

Attributes

logger
task_to_dataset_map

Functions

get_target_preprocessed(→ str)
get_target_overwrite(→ Tuple[str, str])	@return preprocessed target, target_overwrite
get_metadata_filtered_path(→ str)
get_metadata_filtered(→ List[Dict[str, Any]])
save_metadata_filtered(task, metadata_filtered[, ...])
exists_metadata_filtered(→ bool)
get_attribute_for_entity(→ Any)
get_unlearned_model_folder(→ str)
exists_unlearned_model(→ bool)
get_generated_dataset_folder(→ str)
get_generated_dataset_file(→ str)
exists_unlearned_dataset(→ bool)	Return True if the entity dataset folder contains all expected on_* images.
get_baseline_dataset_folder(→ str)	Return the folder path for method-agnostic baseline (lora_state='off') images.
get_off_image_path(→ str)	Return the path to a baseline (lora_state='off') image for a given entity/seed/prompt.
get_similarity_clip_path(→ str)
get_similarity_clip_df(→ pandas.DataFrame)
calculate_similarity_clip(→ None)
plot_heatmap(df[, figsize, cmap, title])	Plot a heatmap for a square DataFrame with all labels visible.

Module Contents

vision_unlearning.datasets.testbed.logger

vision_unlearning.datasets.testbed.get_target_preprocessed(task: Literal['scenes', 'objects', 'breeds', 'people'], target: str) → str

vision_unlearning.datasets.testbed.get_target_overwrite(task: Literal['scenes', 'objects', 'breeds', 'people'], method: Literal['munba', 'uce', 'distil'], target: str) → Tuple[str, str]

@return preprocessed target, target_overwrite

vision_unlearning.datasets.testbed.get_metadata_filtered_path(task: Literal['scenes', 'objects', 'breeds', 'people'], base_folder: str = 'assets') → str

vision_unlearning.datasets.testbed.get_metadata_filtered(task: Literal['scenes', 'objects', 'breeds', 'people'], base_folder: str = 'assets') → List[Dict[str, Any]]

vision_unlearning.datasets.testbed.save_metadata_filtered(task: Literal['scenes', 'objects', 'breeds', 'people'], metadata_filtered: List[Dict[str, Any]], base_folder: str = 'assets')

vision_unlearning.datasets.testbed.exists_metadata_filtered(task: Literal['scenes', 'objects', 'breeds', 'people'], base_folder: str = 'assets') → bool

vision_unlearning.datasets.testbed.get_attribute_for_entity(metadata_filtered: List[Dict[str, Any]], entity_name: str, attribute: str) → Any

vision_unlearning.datasets.testbed.task_to_dataset_map: Dict[Literal['scenes', 'objects', 'breeds', 'people'], str]

vision_unlearning.datasets.testbed.get_unlearned_model_folder(task: Literal['scenes', 'objects', 'breeds', 'people'], method: Literal['munba', 'uce', 'distil'], num_train_epochs: int, target: str, base_folder: str = 'assets') → str

vision_unlearning.datasets.testbed.exists_unlearned_model(task: Literal['scenes', 'objects', 'breeds', 'people'], method: Literal['munba', 'uce', 'distil'], num_train_epochs: int, target: str, base_folder: str = 'assets') → bool

vision_unlearning.datasets.testbed.get_generated_dataset_folder(task: Literal['scenes', 'objects', 'breeds', 'people'], method: Literal['munba', 'uce', 'distil'], num_train_epochs: int, target: str, base_folder: str = 'assets') → str

vision_unlearning.datasets.testbed.get_generated_dataset_file(lora_state: Literal['on', 'off'], seed: int, prompt: str) → str

vision_unlearning.datasets.testbed.exists_unlearned_dataset(generated_dataset_output_path: str, generate_dataset_seeds: List[int], prompts: List[str]) → bool

Return True if the entity dataset folder contains all expected on_* images.

Entity folders now contain only lora_state=’on’ (unlearned model) images. Baseline lora_state=’off’ images live in the separate baseline folder; see get_baseline_dataset_folder() and get_off_image_path().

Expected file count: len(seeds) * len(prompts) on_*.png files + 1 metadata.jsonl.

vision_unlearning.datasets.testbed.get_baseline_dataset_folder(task: Literal['scenes', 'objects', 'breeds', 'people'], target: str, base_folder: str = 'assets') → str

Return the folder path for method-agnostic baseline (lora_state=’off’) images.

Baseline images are generated once per entity by 0_generate_dataset_original.py and shared across all methods. The folder is distinct from the per-method entity folder returned by get_generated_dataset_folder().

Convention: assets/datasets/generated_{task}_baseline_{target}/

vision_unlearning.datasets.testbed.get_off_image_path(task: Literal['scenes', 'objects', 'breeds', 'people'], target: str, method: Literal['munba', 'uce', 'distil'], num_train_epochs: int, seed: int, prompt: str, base_folder: str = 'assets') → str

Return the path to a baseline (lora_state=’off’) image for a given entity/seed/prompt.

Encapsulates backward-compatibility fallback logic in a single place: 1. If the baseline folder (get_baseline_dataset_folder) exists on disk, use it. 2. Otherwise fall back to the old entity folder (get_generated_dataset_folder),

which was the pre-refactor location for both on_* and off_* images.

This means existing datasets that contain off_* files in the entity folder continue to work transparently until baseline folders are generated.

vision_unlearning.datasets.testbed.get_similarity_clip_path(task: Literal['scenes', 'objects', 'breeds', 'people'], base_folder: str = 'assets') → str

vision_unlearning.datasets.testbed.get_similarity_clip_df(task: Literal['scenes', 'objects', 'breeds', 'people'], base_folder: str = 'assets') → pandas.DataFrame

vision_unlearning.datasets.testbed.calculate_similarity_clip(task: Literal['scenes', 'objects', 'breeds', 'people'], labels: List[str], base_folder: str = 'assets') → None

vision_unlearning.datasets.testbed.plot_heatmap(df, figsize=None, cmap='viridis', title='Heatmap')

Plot a heatmap for a square DataFrame with all labels visible.

Parameters:

• df (pd.DataFrame) – A square DataFrame with same string labels for index and columns.

• figsize (tuple) – Figure size (width, height). Increase if labels overlap.

• cmap (str) – Colormap name for matplotlib.