sparse_caption.coco_caption package

Subpackages

Submodules

sparse_caption.coco_caption.eval module

Created on 31 Mar 2020 23:02:57

@author: jiahuei

sparse_caption.coco_caption.eval.evaluate_caption_json(res_file, ann_file)

Given file paths to the caption prediction and annotation JSON files, will compute captioning metric scores as a dict.

The file paths can be absolute or relative. If relative, the files are assumed to be in coco_caption/results and coco_caption/annotations respectively.

Parameters

  • res_file – File path to the caption prediction JSON file.

  • ann_file – File path to the caption annotation JSON file.

Returns

{

‘Bleu_1’: x.x, ‘Bleu_2’: x.x, ‘Bleu_3’: x.x, ‘Bleu_4’: x.x, ‘METEOR’: x.x, ‘ROUGE_L’: x.x, ‘CIDEr’: x.x, ‘SPICE’: x.x, }

A list of scores for each image:
[

{ ‘image_id’: 2090545563, ‘Bleu_1’: x.x, ‘Bleu_2’: x.x, ‘Bleu_3’: x.x, ‘Bleu_4’: x.x, ‘METEOR’: x.x, ‘ROUGE_L’: 0.0, ‘CIDEr’: 0.0, ‘SPICE’:

{ ‘All’: {‘pr’: 0.0, ‘re’: 0.0, ‘f’: 0.0, ‘fn’: 24.0, ‘numImages’: 1.0, ‘fp’: 7.0, ‘tp’: 0.0}, ‘Relation’: {‘pr’: 0.0, ‘re’: 0.0, ‘f’: 0.0, ‘fn’: 7.0, ‘numImages’: 1.0, ‘fp’: 0.0, ‘tp’: 0.0}, ‘Cardinality’: {‘pr’: 0.0, ‘re’: 0.0, ‘f’: 0.0, ‘fn’: 3.0, ‘numImages’: 1.0, ‘fp’: 0.0, ‘tp’: 0.0}, ‘Attribute’: {‘pr’: 0.0, ‘re’: 0.0, ‘f’: 0.0, ‘fn’: 7.0, ‘numImages’: 1.0, ‘fp’: 5.0, ‘tp’: 0.0}, ‘Size’: {‘pr’: 0.0, ‘re’: 0.0, ‘f’: 0.0, ‘fn’: 1.0, ‘numImages’: 1.0, ‘fp’: 0.0, ‘tp’: 0.0}, ‘Color’: {‘pr’: 0.0, ‘re’: 0.0, ‘f’: 0.0, ‘fn’: 1.0, ‘numImages’: 1.0, ‘fp’: 0.0, ‘tp’: 0.0}, ‘Object’: {‘pr’: 0.0, ‘re’: 0.0, ‘f’: 0.0, ‘fn’: 10.0, ‘numImages’: 1.0, ‘fp’: 2.0, ‘tp’: 0.0} }

]

COCO Eval object

Return type

A dict with scores, with the following structure

Module contents

Created on 09 Jul 2020 23:36:50 @author: jiahuei