|
|
|
|
|
|
|
|
|
|
|
model:
|
|
arch: alpro_qa
|
|
num_classes: 1500
|
|
|
|
load_finetuned: True
|
|
|
|
finetuned: "https://storage.googleapis.com/sfr-vision-language-research/LAVIS/models/ALPRO/alpro_msrvtt_qa.pth"
|
|
pretrained: "https://storage.googleapis.com/sfr-vision-language-research/LAVIS/models/ALPRO/alpro_pretrain.pt"
|
|
|
|
timesformer:
|
|
n_frms: 16
|
|
image_size: 224
|
|
|
|
patch_size: 16
|
|
attn_drop_rate: 0.
|
|
drop_rate: 0.
|
|
drop_path_rate: 0.1
|
|
|
|
use_grad_ckpt: True
|
|
ckpt_layer: 12
|
|
|
|
|
|
med_config_path: "configs/models/bert_config_alpro.json"
|
|
|
|
preprocess:
|
|
vis_processor:
|
|
train:
|
|
name: "alpro_video_train"
|
|
n_frms: 16
|
|
image_size: 224
|
|
eval:
|
|
name: "alpro_video_eval"
|
|
n_frms: 16
|
|
image_size: 224
|
|
text_processor:
|
|
train:
|
|
name: "blip_caption"
|
|
eval:
|
|
name: "blip_caption"
|
|
|