File size: 3,986 Bytes
8bc3305 | 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 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 | '''
# author: Zhiyuan Yan
# email: zhiyuanyan@link.cuhk.edu.cn
# date: 2023-0706
# description: Class for the CLIPDetector
Functions in the Class are summarized as:
1. __init__: Initialization
2. build_backbone: Backbone-building
3. build_loss: Loss-function-building
4. features: Feature-extraction
5. classifier: Classification
6. get_losses: Loss-computation
7. get_train_metrics: Training-metrics-computation
8. get_test_metrics: Testing-metrics-computation
9. forward: Forward-propagation
Reference:
@inproceedings{radford2021learning,
title={Learning transferable visual models from natural language supervision},
author={Radford, Alec and Kim, Jong Wook and Hallacy, Chris and Ramesh, Aditya and Goh, Gabriel and Agarwal, Sandhini and Sastry, Girish and Askell, Amanda and Mishkin, Pamela and Clark, Jack and others},
booktitle={International conference on machine learning},
pages={8748--8763},
year={2021},
organization={PMLR}
}
'''
import os
import datetime
import logging
import numpy as np
from sklearn import metrics
from typing import Union
from collections import defaultdict
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
from torch.nn import DataParallel
from torch.utils.tensorboard import SummaryWriter
from metrics.base_metrics_class import calculate_metrics_for_train
from .base_detector import AbstractDetector
from detectors import DETECTOR
from networks import BACKBONE
from loss import LOSSFUNC
from transformers import AutoProcessor, CLIPModel, ViTModel, ViTConfig
import loralib as lora
import copy
logger = logging.getLogger(__name__)
@DETECTOR.register_module(module_name='clip')
class CLIPDetector(AbstractDetector):
def __init__(self, config):
super().__init__()
self.config = config
self.backbone = self.build_backbone(config)
self.head = nn.Linear(768, 2)
self.loss_func = self.build_loss(config)
def build_backbone(self, config):
# prepare the backbone
_, backbone = get_clip_visual(model_name=config['pretrained'])
return backbone
def build_loss(self, config):
# prepare the loss function
loss_class = LOSSFUNC[config['loss_func']]
loss_func = loss_class()
return loss_func
def features(self, data_dict: dict) -> torch.tensor:
feat = self.backbone(data_dict['image'])['pooler_output']
return feat
def classifier(self, features: torch.tensor) -> torch.tensor:
return self.head(features)
def get_losses(self, data_dict: dict, pred_dict: dict) -> dict:
label = data_dict['label']
pred = pred_dict['cls']
loss = self.loss_func(pred, label)
loss_dict = {'overall': loss}
return loss_dict
def get_train_metrics(self, data_dict: dict, pred_dict: dict) -> dict:
label = data_dict['label']
pred = pred_dict['cls']
# compute metrics for batch data
auc, eer, acc, ap = calculate_metrics_for_train(label.detach(), pred.detach())
metric_batch_dict = {'acc': acc, 'auc': auc, 'eer': eer, 'ap': ap}
return metric_batch_dict
def forward(self, data_dict: dict, inference=False) -> dict:
# get the features by backbone
features = self.features(data_dict)
# get the prediction by classifier
pred = self.classifier(features)
# get the probability of the pred
prob = torch.softmax(pred, dim=1)[:, 1]
# build the prediction dict for each output
pred_dict = {'cls': pred, 'prob': prob, 'feat': features}
return pred_dict
def get_clip_visual(model_name = "openai/clip-vit-base-patch16"):
processor = AutoProcessor.from_pretrained(model_name)
model = CLIPModel.from_pretrained(model_name)
return processor, model.vision_model
|