Spaces:
Running
on
Zero
Running
on
Zero
| import torch | |
| import torch.nn as nn | |
| from transformers import ViTMAEForPreTraining, AutoConfig, AutoImageProcessor | |
| class MAEVisionTower(nn.Module): | |
| def __init__(self, vision_tower, args, cache_dir='./cache_dir', delay_load=False): | |
| super().__init__() | |
| self.is_loaded = False | |
| self.cache_dir = cache_dir | |
| self.vision_tower_name = vision_tower | |
| self.select_layer = args.mm_vision_select_layer | |
| self.select_feature = getattr(args, 'mm_vision_select_feature', 'patch') | |
| if not delay_load: | |
| self.load_model() | |
| else: | |
| self.cfg_only = AutoConfig.from_pretrained(self.vision_tower_name, cache_dir=self.cache_dir) | |
| def load_model(self): | |
| self.image_processor = AutoImageProcessor.from_pretrained(self.vision_tower_name, cache_dir=self.cache_dir) | |
| vision_tower = ViTMAEForPreTraining.from_pretrained(self.vision_tower_name, cache_dir=self.cache_dir) | |
| self.vision_tower = vision_tower.vit | |
| self.vision_tower.requires_grad_(False) | |
| self.is_loaded = True | |
| def feature_select(self, image_forward_outs): | |
| image_features = image_forward_outs.hidden_states[self.select_layer] | |
| if self.select_feature == 'patch': | |
| image_features = image_features[:, 1:] | |
| elif self.select_feature == 'cls_patch': | |
| image_features = image_features | |
| else: | |
| raise ValueError(f'Unexpected select feature: {self.select_feature}') | |
| # print(image_features.shape) | |
| return image_features | |
| def forward(self, images): | |
| if type(images) is list: | |
| image_features = [] | |
| for image in images: | |
| image_forward_out = self.vision_tower(image.to(device=self.device, dtype=self.dtype).unsqueeze(0), output_hidden_states=True) | |
| image_feature = self.feature_select(image_forward_out).to(image.dtype) | |
| image_features.append(image_feature) | |
| else: | |
| image_forward_outs = self.vision_tower(images.to(device=self.device, dtype=self.dtype), output_hidden_states=True) | |
| image_features = self.feature_select(image_forward_outs).to(images.dtype) | |
| return image_features | |
| def dummy_feature(self): | |
| return torch.zeros(1, self.hidden_size, device=self.device, dtype=self.dtype) | |
| def dtype(self): | |
| return self.vision_tower.dtype | |
| def device(self): | |
| return self.vision_tower.device | |
| def config(self): | |
| if self.is_loaded: | |
| return self.vision_tower.config | |
| else: | |
| return self.cfg_only | |
| def hidden_size(self): | |
| return self.config.hidden_size | |
| def num_patches(self): | |
| return (self.config.image_size // self.config.patch_size) ** 2 | |