关于gradcam在异常检测场景的应用问题

[EricLee0224]

作者您好，
首先感谢您的代码贡献，非常简洁，关键注释非常清晰！已按照readme已经成功跑通示例~目前希望依托您的代码框架，进一步想试一试引入自己的预训练网络，生成gradcam，进行图像异常检测。现有一个基于Res18预训练模型，前面添加了head，后面添加了几层额外的卷积层和fc层（最终输出分别是0：正常、1：异常），对自己的正常数据集进行无监督学习训练得到权重，用于对异常图片进行异常检测。然后利用gradcam在异常图像上标注出异常的位置。
现有问题是如何将前面提到的自己的模型权重引入框架？自己试了试之后会有如下报错，请问可能是什么问题？

(gradcam)_____@server3090-X570-AORUS-PRO-WIFI:~/Grad-CAM.pytorch-master$ python main.py
feature shape:torch.Size([1, 512, 7, 7])
/home/____/.conda/envs/gradcam/lib/python3.8/site-packages/torch/nn/modules/module.py:1033: UserWarning: Using a non-full backward hook when the forward contains multiple autograd Nodes is deprecated and will be removed in future versions. This hook will be missing some grad_input. Please use register_full_backward_hook to get the documented behavior.
warnings.warn("Using a non-full backward hook when the forward contains multiple autograd Nodes "
feature shape:torch.Size([1, 512, 7, 7])

ps.最终确实能够生成图，但明显不是基于我自己的模型来生成的。
以下是自己修改后的main.py：

-- coding: utf-8 --

import argparse
import os
import re

import torch.nn as nn
import torch.nn.functional as F
from torchvision.models import resnet18
import cv2
import numpy as np
import torch
from skimage import io
from torch import nn
from torchvision import models

from interpretability.grad_cam import GradCAM, GradCamPlusPlus
from interpretability.guided_back_propagation import GuidedBackPropagation

def get_net(net_name, weight_path=None):
"""
根据名称获取模型
:param net_name: 网络名称
:param weight_path: 与训练权重路径
:return:
"""
pretrain = weight_path is None # 没有指定权重路径，加载默认的预训练权重
if net_name in ['vgg', 'vgg16']:
net = models.vgg16(pretrained=pretrain)
elif net_name in ['resnet', 'resnet18']:
net = models.resnet18(pretrained=pretrain)
else:
raise ValueError('invalid network name:{}'.format(net_name))
# 加载指定路径的权重参数
if weight_path is not None and net_name.startswith('densenet'):
pattern = re.compile(
r'^(.*denselayer\d+.(?:norm|relu|conv)).((?:[12]).(?:weight|bias|running_mean|running_var))$')
state_dict = torch.load(weight_path)
for key in list(state_dict.keys()):
res = pattern.match(key)
if res:
new_key = res.group(1) + res.group(2)
state_dict[new_key] = state_dict[key]
del state_dict[key]
net.load_state_dict(state_dict)
elif weight_path is not None:
net.load_state_dict({k.replace('resnet18.',''):v for k,v in torch.load(weight_path).items()},strict=False)
return net

def get_last_conv_name(net):
"""
获取网络的最后一个卷积层的名字
:param net:
:return:
"""
layer_name = None
for name, m in net.named_modules():
if isinstance(m, nn.Conv2d):
layer_name = name
return layer_name

def prepare_input(image):
image = image.copy()

# 归一化
#means = np.array([0.485, 0.456, 0.406])
#stds = np.array([0.229, 0.224, 0.225])
#image -= means
#image /= stds

image = np.ascontiguousarray(np.transpose(image, (2, 0, 1)))  # channel first
image = image[np.newaxis, ...]  # 增加batch维

return torch.tensor(image, requires_grad=True)

def gen_cam(image, mask):
"""
生成CAM图
:param image: [H,W,C],原始图像
:param mask: [H,W],范围0~1
:return: tuple(cam,heatmap)
"""
# mask转为heatmap
heatmap = cv2.applyColorMap(np.uint8(255 * mask), cv2.COLORMAP_JET)
heatmap = np.float32(heatmap) / 255
heatmap = heatmap[..., ::-1] # gbr to rgb

# 合并heatmap到原始图像
cam = heatmap + np.float32(image)
return norm_image(cam), (heatmap * 255).astype(np.uint8)

def norm_image(image):
"""
标准化图像
:param image: [H,W,C]
:return:
"""
image = image.copy()
image -= np.max(np.min(image), 0)
image /= np.max(image)
image *= 255.
return np.uint8(image)

def gen_gb(grad):
"""
生guided back propagation 输入图像的梯度
:param grad: tensor,[3,H,W]
:return:
"""
# 标准化
grad = grad.data.numpy()
gb = np.transpose(grad, (1, 2, 0))
return gb

def save_image(image_dicts, input_image_name, network, output_dir):
prefix = os.path.splitext(input_image_name)[0]
for key, image in image_dicts.items():
io.imsave(os.path.join(output_dir, '{}-{}-{}.jpg'.format(prefix, network, key)), image)

def main(args):
# 输入
img = io.imread(args.image_path)
img = cv2.cvtColor(img, cv2.COLOR_GRAY2BGR)
img = np.float32(cv2.resize(img, (224, 224))) / 255
inputs = prepare_input(img)
# 输出图像
image_dict = {}
# 网络
net = get_net(args.network, args.weight_path)
# Grad-CAM
layer_name = get_last_conv_name(net) if args.layer_name is None else args.layer_name
grad_cam = GradCAM(net, layer_name)
mask = grad_cam(inputs, args.class_id) # cam mask
image_dict['cam'], image_dict['heatmap'] = gen_cam(img, mask)
grad_cam.remove_handlers()
# Grad-CAM++
grad_cam_plus_plus = GradCamPlusPlus(net, layer_name)
mask_plus_plus = grad_cam_plus_plus(inputs, args.class_id) # cam mask
image_dict['cam++'], image_dict['heatmap++'] = gen_cam(img, mask_plus_plus)
grad_cam_plus_plus.remove_handlers()

# GuidedBackPropagation
gbp = GuidedBackPropagation(net)
inputs.grad.zero_()  # 梯度置零
grad = gbp(inputs)

gb = gen_gb(grad)
image_dict['gb'] = norm_image(gb)
# 生成Guided Grad-CAM
cam_gb = gb * mask[..., np.newaxis]
image_dict['cam_gb'] = norm_image(cam_gb)

save_image(image_dict, os.path.basename(args.image_path), args.network, args.output_dir)

if name == 'main':
parser = argparse.ArgumentParser()
parser.add_argument('--network', type=str, default='resnet18',
help='ImageNet classification network')
parser.add_argument('--image-path', type=str, default='./Cutpaste_examples/icecream/XGQK_test.jpg',
help='input image path')
parser.add_argument('--weight-path', type=str, default='./Cutpaste_examples/icecream/model-icecream-cutpaste-normal.pth',
help='weight path of the model')
parser.add_argument('--layer-name', type=str, default=None,
help='last convolutional layer name')
parser.add_argument('--class-id', type=int, default=None,
help='class id')
parser.add_argument('--output-dir', type=str, default='results',
help='output directory to save results')
arguments = parser.parse_args()

main(arguments)