File size: 1,517 Bytes
256a159 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 |
from opencompass.multimodal.models.llava import LLaVABasePromptConstructor, LLaVABasePostProcessor
# dataloader settings
val_pipeline = [
dict(type='mmpretrain.LoadImageFromFile'),
dict(type='mmpretrain.ToPIL', to_rgb=True),
dict(type='mmpretrain.torchvision/Resize',
size=(224, 224),
interpolation=3),
dict(type='mmpretrain.torchvision/ToTensor'),
dict(
type='mmpretrain.torchvision/Normalize',
mean=(0.48145466, 0.4578275, 0.40821073),
std=(0.26862954, 0.26130258, 0.27577711),
),
dict(type='mmpretrain.PackInputs', algorithm_keys=['image_id']),
]
dataset = dict(type='mmpretrain.Flickr30kCaption',
data_root='data/flickr30k',
ann_file='annotations/dataset_flickr30k.json',
data_prefix='images',
split='val',
pipeline=val_pipeline)
llava_flickr30k_dataloader = dict(
batch_size=1,
num_workers=4,
dataset=dataset,
collate_fn=dict(type='pseudo_collate'),
sampler=dict(type='DefaultSampler', shuffle=False),
)
# model settings
llava_flickr30k_model = dict(
type='llava',
model_path='/path/to/llava',
is_caption_task=True,
prompt_constructor=dict(type=LLaVABasePromptConstructor),
post_processor=dict(type=LLaVABasePostProcessor)
) # noqa
# evaluation settings
llava_flickr30k_evaluator = [
dict(
type='mmpretrain.COCOCaption',
ann_file='data/flickr30k/annotations/flickr30k_val_gt.json',
) # noqa
]
|