Function decode_text

fourm/utils/plotting_utils.py:473–510

Decodes a text sequence from a model dictionary. Args: mod_dict (dict): Model output dictionary. key (str): Key of the text modality to decode. text_tokenizer (tokenizers.Tokenizer): Text tokenizer.

(mod_dict, key, text_tokenizer)

    return all_metadata

def decode_text(mod_dict, key, text_tokenizer):
    """
    Decodes a text sequence from a model dictionary.

    Args:
        mod_dict (dict): Model output dictionary.
        key (str): Key of the text modality to decode.
        text_tokenizer (tokenizers.Tokenizer): Text tokenizer.
    """
    input_texts, target_texts, merged_texts = [], [], []

    sentinel_ids = set(get_sentinel_to_id_mapping(text_tokenizer).values())
    B = mod_dict[key]['tensor'].shape[0]

    for i in range(B):

        input_seq = mod_dict[key]['tensor'][i]
        input_seq = input_seq[mod_dict[key]['input_mask'][i] == 0]
        input_seq = input_seq.tolist()
        
        target_seq = mod_dict[key]['tensor'][i]
        target_seq = target_seq[mod_dict[key]['target_mask'][i] == 0]
        target_seq = target_seq.tolist()
        
        merged_seq = merge_span_masking(input_seq, target_seq, sentinel_ids=sentinel_ids)
    
        input_text = text_tokenizer.decode(input_seq, skip_special_tokens=False)
        target_text = text_tokenizer.decode(target_seq, skip_special_tokens=False)
        merged_text = text_tokenizer.decode(merged_seq, skip_special_tokens=False)

        input_texts.append(input_text)
        target_texts.append(target_text)
        merged_texts.append(merged_text)

    if B == 1:
        input_texts, target_texts, merged_texts = input_texts[0], target_texts[0], merged_texts[0]
    
    return input_texts, target_texts, merged_texts

def decode_sam_instances(mod_dict, tokenizers, text_tokenizer, key='sam_instance', image_size=224, token_len=16):
    '''