''' * Tag2Text * Written by Xinyu Huang ''' import warnings warnings.filterwarnings("ignore") from .vit import VisionTransformer, interpolate_pos_embed from .swin_transformer import SwinTransformer, interpolate_relative_pos_embed from .med import BertConfig, BertModel, BertLMHeadModel from .utils import tra_array from transformers import BertTokenizer import torch from torch import nn import torch.nn.functional as F import os from urllib.parse import urlparse from timm.models.hub import download_cached_file import json import math import numpy as np def read_json(rpath): with open(rpath, 'r') as f: return json.load(f) # delete some tags that may disturb captioning # 127: "quarter"; 2961: "back"; 3351: "two"; 3265: "three"; 3338: "four"; 3355: "five"; 3359: "one" delete_tag_index = [127,2961, 3351, 3265, 3338, 3355, 3359] # adjust thresholds for some tags # default threshold: 0.68 # 2701: "person"; 2828: "man"; 1167: "woman"; tag_thrshold = {2701:0.7, 2828: 0.7, 1167: 0.7} class Tag2Text_Caption(nn.Module): def __init__(self, med_config = 'configs/med_config.json', image_size = 384, vit = 'base', vit_grad_ckpt = False, vit_ckpt_layer = 0, prompt = 'a picture of ', threshold = 0.68, ): """ Args: med_config (str): path for the mixture of encoder-decoder model's configuration file image_size (int): input image size vit (str): model size of vision transformer """ super().__init__() if vit=='swin_b': if image_size == 224: vision_config_path = 'configs/swin/config_swinB_224.json' elif image_size == 384: vision_config_path = 'configs/swin/config_swinB_384.json' vision_config = read_json(vision_config_path) assert image_size == vision_config['image_res'] # assert config['patch_size'] == 32 vision_width = vision_config['vision_width'] self.visual_encoder = SwinTransformer(img_size=vision_config['image_res'], patch_size=4, in_chans=3, embed_dim=vision_config['embed_dim'], depths=vision_config['depths'], num_heads=vision_config['num_heads'], window_size=vision_config['window_size'], mlp_ratio=4., qkv_bias=True, drop_rate=0.0, drop_path_rate=0.1, ape=False, patch_norm=True, use_checkpoint=False) else: self.visual_encoder, vision_width = create_vit(vit,image_size, vit_grad_ckpt, vit_ckpt_layer) self.tokenizer = init_tokenizer() # create the decoder decoder_config = BertConfig.from_json_file(med_config) decoder_config.encoder_width = 768 self.text_decoder = BertLMHeadModel(config=decoder_config) # create encoder encoder_config = BertConfig.from_json_file(med_config) encoder_config.encoder_width = vision_width self.tag_encoder = BertModel(config=encoder_config, add_pooling_layer=False) self.prompt = prompt self.prompt_length = len(self.tokenizer(self.prompt).input_ids)-1 self.threshold = threshold num_features = 768 self.num_class = 3429 q2l_config = BertConfig.from_json_file('configs/q2l_config.json') q2l_config.encoder_width = vision_width self.vision_multi = BertModel(config=q2l_config, add_pooling_layer=False) self.vision_multi.resize_token_embeddings(len(self.tokenizer)) self.label_embed = nn.Embedding(self.num_class, q2l_config.hidden_size) self.fc = GroupWiseLinear(self.num_class, num_features, bias=True) self.del_selfattention() tie_encoder_decoder_weights(self.tag_encoder,self.vision_multi,'',' ') self.tag_array = tra_array self.class_threshold = torch.ones(self.num_class) * self.threshold for key,value in tag_thrshold.items(): self.class_threshold[key] = value def del_selfattention(self): del self.vision_multi.embeddings for layer in self.vision_multi.encoder.layer: del layer.attention def generate(self, image, sample=False, num_beams=3, max_length=30, min_length=10, top_p=0.9, repetition_penalty=1.0, tag_input = None, return_tag_predict = False): image_embeds = self.visual_encoder(image) image_atts = torch.ones(image_embeds.size()[:-1],dtype=torch.long).to(image.device) #==============generate tag==============# if tag_input == None: image_spatial_embeds = image_embeds[:,1:,:] image_cls_embeds = image_embeds[:,0,:] bs = image_spatial_embeds.shape[0] label_embed = self.label_embed.weight.unsqueeze(0).repeat(bs,1,1) mlr_tagembedding = self.vision_multi(encoder_embeds = label_embed, encoder_hidden_states = image_embeds, encoder_attention_mask = image_atts, return_dict = False, mode = 'mlr', ) logits = self.fc(mlr_tagembedding[0]) # targets = torch.where(torch.sigmoid(logits) > self.threshold , torch.tensor(1.0).to(image.device), torch.zeros(self.num_class).to(image.device)) targets = torch.where(torch.sigmoid(logits) > self.class_threshold.to(image.device) , torch.tensor(1.0).to(image.device), torch.zeros(self.num_class).to(image.device)) tag = targets.cpu().numpy() tag[:,delete_tag_index] = 0 bs = image.size(0) tag_input = [] for b in range(bs): index = np.argwhere(tag[b] == 1) token = self.tag_array[index].squeeze(axis = 1) tag_input.append(' | '.join(token)) #========================================# if not sample: image_embeds = image_embeds.repeat_interleave(num_beams,dim=0) image_atts = image_atts.repeat_interleave(num_beams,dim=0) tag_input_temp = [] for tag in tag_input: for i in range(num_beams): tag_input_temp.append(tag) tag_input = tag_input_temp tag_input_tokenzier = self.tokenizer(tag_input, padding='max_length', truncation=True, max_length=40, return_tensors="pt").to(image.device) encoder_input_ids = tag_input_tokenzier.input_ids encoder_input_ids[:,0] = self.tokenizer.enc_token_id output_tagembedding = self.tag_encoder(encoder_input_ids, attention_mask = tag_input_tokenzier.attention_mask, encoder_hidden_states = image_embeds, encoder_attention_mask = image_atts, return_dict = True, ) prompt = [self.prompt] * image.size(0) input_ids = self.tokenizer(prompt, return_tensors="pt").input_ids.to(image.device) input_ids[:,0] = self.tokenizer.bos_token_id input_ids = input_ids[:, :-1] if sample: #nucleus sampling model_kwargs = {"encoder_hidden_states": output_tagembedding.last_hidden_state, "encoder_attention_mask":None} outputs = self.text_decoder.generate(input_ids=input_ids, max_length=max_length, min_length=min_length, do_sample=True, top_p=top_p, num_return_sequences=1, eos_token_id=self.tokenizer.sep_token_id, pad_token_id=self.tokenizer.pad_token_id, repetition_penalty=1.1, **model_kwargs) else: #beam search model_kwargs = {"encoder_hidden_states": output_tagembedding.last_hidden_state, "encoder_attention_mask":None} outputs = self.text_decoder.generate(input_ids=input_ids, max_length=max_length, min_length=min_length, num_beams=num_beams, eos_token_id=self.tokenizer.sep_token_id, pad_token_id=self.tokenizer.pad_token_id, repetition_penalty=repetition_penalty, **model_kwargs) captions = [] for output in outputs: caption = self.tokenizer.decode(output, skip_special_tokens=True) captions.append(caption[len(self.prompt):]) if return_tag_predict == True: if sample: return captions, tag_input else: return captions, tag_input[0:int(len(tag_input)/num_beams)] return captions def tag2text_caption(pretrained='',**kwargs): model = Tag2Text_Caption(**kwargs) if pretrained: if kwargs['vit'] == 'swin_b': model,msg = load_checkpoint_swinbase(model,pretrained,kwargs) else: model,msg = load_checkpoint(model,pretrained) # print('vit:',kwargs['vit']) # print('msg_v2',msg) return model from typing import List def tie_encoder_decoder_weights(encoder: nn.Module, decoder: nn.Module, base_model_prefix: str, skip_key:str): uninitialized_encoder_weights: List[str] = [] if decoder.__class__ != encoder.__class__: logger.info( f"{decoder.__class__} and {encoder.__class__} are not equal. In this case make sure that all encoder weights are correctly initialized." ) def tie_encoder_to_decoder_recursively( decoder_pointer: nn.Module, encoder_pointer: nn.Module, module_name: str, uninitialized_encoder_weights: List[str], skip_key: str, depth=0, ): assert isinstance(decoder_pointer, nn.Module) and isinstance( encoder_pointer, nn.Module ), f"{decoder_pointer} and {encoder_pointer} have to be of type torch.nn.Module" if hasattr(decoder_pointer, "weight") and skip_key not in module_name: assert hasattr(encoder_pointer, "weight") encoder_pointer.weight = decoder_pointer.weight if hasattr(decoder_pointer, "bias"): assert hasattr(encoder_pointer, "bias") encoder_pointer.bias = decoder_pointer.bias # print(module_name+' is tied') return encoder_modules = encoder_pointer._modules decoder_modules = decoder_pointer._modules if len(decoder_modules) > 0: assert ( len(encoder_modules) > 0 ), f"Encoder module {encoder_pointer} does not match decoder module {decoder_pointer}" all_encoder_weights = set([module_name + "/" + sub_name for sub_name in encoder_modules.keys()]) encoder_layer_pos = 0 for name, module in decoder_modules.items(): if name.isdigit(): encoder_name = str(int(name) + encoder_layer_pos) decoder_name = name if not isinstance(decoder_modules[decoder_name], type(encoder_modules[encoder_name])) and len( encoder_modules ) != len(decoder_modules): # this can happen if the name corresponds to the position in a list module list of layers # in this case the decoder has added a cross-attention that the encoder does not have # thus skip this step and subtract one layer pos from encoder encoder_layer_pos -= 1 continue elif name not in encoder_modules: continue elif depth > 500: raise ValueError( "Max depth of recursive function `tie_encoder_to_decoder` reached. It seems that there is a circular dependency between two or more `nn.Modules` of your model." ) else: decoder_name = encoder_name = name tie_encoder_to_decoder_recursively( decoder_modules[decoder_name], encoder_modules[encoder_name], module_name + "/" + name, uninitialized_encoder_weights, skip_key, depth=depth + 1, ) all_encoder_weights.remove(module_name + "/" + encoder_name) uninitialized_encoder_weights += list(all_encoder_weights) # tie weights recursively tie_encoder_to_decoder_recursively(decoder, encoder, base_model_prefix, uninitialized_encoder_weights, skip_key) class GroupWiseLinear(nn.Module): # could be changed to: # output = torch.einsum('ijk,zjk->ij', x, self.W) # or output = torch.einsum('ijk,jk->ij', x, self.W[0]) def __init__(self, num_class, hidden_dim, bias=True): super().__init__() self.num_class = num_class self.hidden_dim = hidden_dim self.bias = bias self.W = nn.Parameter(torch.Tensor(1, num_class, hidden_dim)) if bias: self.b = nn.Parameter(torch.Tensor(1, num_class)) self.reset_parameters() def reset_parameters(self): stdv = 1. / math.sqrt(self.W.size(2)) for i in range(self.num_class): self.W[0][i].data.uniform_(-stdv, stdv) if self.bias: for i in range(self.num_class): self.b[0][i].data.uniform_(-stdv, stdv) def forward(self, x): # x: B,K,d x = (self.W * x).sum(-1) if self.bias: x = x + self.b return x def init_tokenizer(): tokenizer = BertTokenizer.from_pretrained('bert-base-uncased') tokenizer.add_special_tokens({'bos_token':'[DEC]'}) tokenizer.add_special_tokens({'additional_special_tokens':['[ENC]']}) tokenizer.enc_token_id = tokenizer.additional_special_tokens_ids[0] return tokenizer def create_vit(vit, image_size, use_grad_checkpointing=False, ckpt_layer=0, drop_path_rate=0): assert vit in ['base', 'large'], "vit parameter must be base or large" if vit=='base': vision_width = 768 visual_encoder = VisionTransformer(img_size=image_size, patch_size=16, embed_dim=vision_width, depth=12, num_heads=12, use_grad_checkpointing=use_grad_checkpointing, ckpt_layer=ckpt_layer, drop_path_rate=0 or drop_path_rate ) elif vit=='large': vision_width = 1024 visual_encoder = VisionTransformer(img_size=image_size, patch_size=16, embed_dim=vision_width, depth=24, num_heads=16, use_grad_checkpointing=use_grad_checkpointing, ckpt_layer=ckpt_layer, drop_path_rate=0.1 or drop_path_rate ) return visual_encoder, vision_width def is_url(url_or_filename): parsed = urlparse(url_or_filename) return parsed.scheme in ("http", "https") def load_checkpoint(model,url_or_filename): if is_url(url_or_filename): cached_file = download_cached_file(url_or_filename, check_hash=False, progress=True) checkpoint = torch.load(cached_file, map_location='cpu') elif os.path.isfile(url_or_filename): checkpoint = torch.load(url_or_filename, map_location='cpu') else: raise RuntimeError('checkpoint url or path is invalid') state_dict = checkpoint['model'] state_dict['visual_encoder.pos_embed'] = interpolate_pos_embed(state_dict['visual_encoder.pos_embed'],model.visual_encoder) if 'visual_encoder_m.pos_embed' in model.state_dict().keys(): state_dict['visual_encoder_m.pos_embed'] = interpolate_pos_embed(state_dict['visual_encoder_m.pos_embed'], model.visual_encoder_m) for key in model.state_dict().keys(): if key in state_dict.keys(): if state_dict[key].shape!=model.state_dict()[key].shape: del state_dict[key] msg = model.load_state_dict(state_dict,strict=False) print('load checkpoint from %s'%url_or_filename) return model,msg def load_checkpoint_swinbase(model,url_or_filename,kwargs): if kwargs['image_size'] == 224: vision_config_path = 'configs/swin/config_swinB_224.json' elif kwargs['image_size'] == 384: vision_config_path = 'configs/swin/config_swinB_384.json' elif kwargs['image_size'] == 480: vision_config_path = 'configs/swin/config_swinB_480.json' elif kwargs['image_size'] == 576: vision_config_path = 'configs/swin/config_swinB_576.json' elif kwargs['image_size'] == 608: vision_config_path = 'configs/swin/config_swinB_608.json' window_size = read_json(vision_config_path)['window_size'] # print('--------------') # print(url_or_filename) # print('--------------') if is_url(url_or_filename): cached_file = download_cached_file(url_or_filename, check_hash=False, progress=True) checkpoint = torch.load(cached_file, map_location='cpu') elif os.path.isfile(url_or_filename): checkpoint = torch.load(url_or_filename, map_location='cpu') else: raise RuntimeError('checkpoint url or path is invalid') state_dict = checkpoint['model'] for k in list(state_dict.keys()): if 'relative_position_bias_table' in k: dst_num_pos = (2 * window_size - 1) ** 2 state_dict[k] = interpolate_relative_pos_embed(state_dict[k], dst_num_pos, param_name=k) elif ('relative_position_index' in k) or ('attn_mask' in k): del state_dict[k] msg = model.load_state_dict(state_dict,strict=False) print('load checkpoint from %s'%url_or_filename) return model,msg