upload files

Inference_OtherMethods/ResNet50_gradcam.py

@@ -0,0 +1,155 @@
+import sys
+sys.path.append('../')
+import torchvision.transforms as transforms
+import torchvision.transforms.functional as F
+import torch
+import torch.nn as nn
+import argparse
+import os
+from tqdm import tqdm
+import time
+import numpy as np
+from torchvision import models, transforms
+import torch.nn.functional as F
+from utils import AverageMeter
+from utils import Metrics
+from utils.LoadData import data_loader
+from utils.cam import GradCAM
+
+os.chdir('../')
+ROOT_DIR = os.getcwd()
+print('Project Root Dir:',ROOT_DIR)
+IMG_DIR = r'/m2/ILSVRC2012_img_val'
+
+# Static paths
+train_list = os.path.join(ROOT_DIR,'datalist', 'ILSVRC', 'VGG16_train.txt')
+test_list = os.path.join(ROOT_DIR,'datalist','ILSVRC', 'Evaluation_2000.txt')
+
+percent = 0
+def get_arguments():
+ parser = argparse.ArgumentParser(description='-')
+ parser.add_argument("--root_dir", type=str, default=ROOT_DIR)
+ parser.add_argument("--img_dir", type=str, default=IMG_DIR)
+ parser.add_argument("--train_list", type=str, default=train_list)
+ parser.add_argument("--test_list", type=str, default=test_list)
+ parser.add_argument("--batch_size", type=int, default=1)
+ parser.add_argument("--input_size", type=int, default=256)
+ parser.add_argument("--crop_size", type=int, default=224)
+ parser.add_argument("--dataset", type=str, default='imagenet')
+ parser.add_argument("--num_classes", type=int, default=1000)
+ parser.add_argument("--arch", type=str,default='-')
+ parser.add_argument("--lr", type=float, default=0.001)
+ parser.add_argument("--num_workers", type=int, default=0)
+ parser.add_argument("--disp_interval", type=int, default=40)
+ parser.add_argument("--snapshot_dir", type=str, default='/m2/gkartzoni/L-CAM/snapshots/temp')
+ parser.add_argument("--resume", type=str, default='True')
+ parser.add_argument("--restore_from", type=str, default=r'/home/xiaolin/.torch/models/vgg16-397923af.pth')
+ parser.add_argument("--global_counter", type=int, default=0)
+ parser.add_argument("--current_epoch", type=int, default=0)
+ parser.add_argument("--percentage", type=float, default=percent)
+ return parser.parse_args()
+
+
+
+current_epoch=0
+model=None
+args = get_arguments()
+
+top1 = AverageMeter()
+top5 = AverageMeter()
+top1.reset()
+top5.reset()
+
+top1_ = AverageMeter()
+top5_ = AverageMeter()
+top1_.reset()
+top5_.reset()
+
+def get_model(args):
+ model = models.resnet50(pretrained=True).eval()
+ model.cuda()
+ return model
+
+if model is None:
+ model = get_model(args)
+
+
+train_loader, val_loader = data_loader(args, test_path=True)
+
+global_counter = 0
+prob = None
+gt = None
+
+y_mask_image = []
+y_image = []
+
+model.cuda()
+
+
+target_layer = model.layer4[2]
+wrapped_model = GradCAM(model, target_layer)
+k = 0
+time_spend = 0
+import matplotlib.image as mpimg 
+
+
+for idx, dat in tqdm(enumerate(val_loader)):
+ img_path, img, label_in = dat
+ im = img
+ global_counter += 1
+ label = label_in
+ img, label = img.cuda(), label.cuda()
+
+ now = time.time()
+ with torch.no_grad():
+ img_n = Metrics.normalize(img)
+ logits = model(img_n)
+ logits0 = logits
+ logits0 = F.softmax(logits0, dim=1)
+
+ prec1_1, prec5_1 = Metrics.accuracy(logits0.cpu().data, label_in.long(), topk=(1,5))
+ y = logits0.cpu().data.numpy()
+ class_1 = logits0.max(1)[-1] #
+
+ index_gt_y = class_1.long().cpu().data.numpy() #
+
+ Y_i_c = logits0.max(1)[0].item()
+ y_image.append(Y_i_c)
+
+ top1.update(prec1_1[0], img.size()[0])
+ top5.update(prec5_1[0], img.size()[0])
+
+ cam_map,_ = wrapped_model(img_n,index_gt_y)
+ cam_map = F.interpolate(cam_map, size=(224,224), mode='bilinear', align_corners=False)
+ cam_map = Metrics.drop_Npercent(cam_map,args.percentage)
+
+
+ cam_map = cam_map.cuda() 
+ image = img
+ mask_image = cam_map * image 
+
+ mask_image = mask_image.cuda()
+ with torch.no_grad():
+ img_n = Metrics.normalize(mask_image)
+ logits = model(img_n)
+ logits0 = logits
+ logits0 = F.softmax(logits0, dim=1)
+
+
+ prec1_1, prec5_1 = Metrics.accuracy(logits0.cpu().data, label_in.long(), topk=(1,5))
+ y = logits0.cpu().data.numpy()
+
+ Y_i_c_ = logits0[:,index_gt_y][0].item()
+ y_mask_image.append(Y_i_c_)
+
+ top1_.update(prec1_1[0], img.size()[0])
+ top5_.update(prec5_1[0], img.size()[0]) 
+
+y_image = np.array(y_image)
+y_mask_image = np.array(y_mask_image)
+
+
+print('Average Drop is: ', Metrics.AD(y_image,y_mask_image))
+print('Increase in confidence is: ', Metrics.IC(y_image,y_mask_image))
+print('Top1:', top1.avg, 'Top5:',top5.avg)
+print('Top1:', top1_.avg, 'Top5:',top5_.avg)

Inference_OtherMethods/ResNet50_gradcapp.py

@@ -0,0 +1,156 @@
+import sys
+sys.path.append('../')
+import torchvision.transforms as transforms
+import torchvision.transforms.functional as F
+import torch
+import torch.nn as nn
+import argparse
+import os
+from tqdm import tqdm
+import time
+import numpy as np
+from torchvision import models, transforms
+import torch.nn.functional as F
+from utils import AverageMeter
+from utils import Metrics
+from utils.LoadData import data_loader
+from utils.cam import GradCAM,GradCAMpp,ScoreCAM,CAM
+
+os.chdir('../')
+ROOT_DIR = os.getcwd()
+print('Project Root Dir:',ROOT_DIR)
+IMG_DIR = r'/m2/ILSVRC2012_img_val'
+
+# Static paths
+train_list = os.path.join(ROOT_DIR,'datalist', 'ILSVRC', 'VGG16_train.txt')
+test_list = os.path.join(ROOT_DIR,'datalist','ILSVRC', 'Evaluation_2000.txt')
+
+percent = 0.85
+def get_arguments():
+ parser = argparse.ArgumentParser(description='-')
+ parser.add_argument("--root_dir", type=str, default=ROOT_DIR)
+ parser.add_argument("--img_dir", type=str, default=IMG_DIR)
+ parser.add_argument("--train_list", type=str, default=train_list)
+ parser.add_argument("--test_list", type=str, default=test_list)
+ parser.add_argument("--batch_size", type=int, default=1)
+ parser.add_argument("--input_size", type=int, default=256)
+ parser.add_argument("--crop_size", type=int, default=224)
+ parser.add_argument("--dataset", type=str, default='imagenet')
+ parser.add_argument("--num_classes", type=int, default=1000)
+ parser.add_argument("--arch", type=str,default='-')
+ parser.add_argument("--lr", type=float, default=0.001)
+ parser.add_argument("--num_workers", type=int, default=0)
+ parser.add_argument("--disp_interval", type=int, default=40)
+ parser.add_argument("--snapshot_dir", type=str, default='/m2/gkartzoni/L-CAM/snapshots/temp')
+ parser.add_argument("--resume", type=str, default='True')
+ parser.add_argument("--restore_from", type=str, default=r'/home/xiaolin/.torch/models/vgg16-397923af.pth')
+ parser.add_argument("--global_counter", type=int, default=0)
+ parser.add_argument("--current_epoch", type=int, default=0)
+ parser.add_argument("--percentage", type=float, default=percent)
+ return parser.parse_args()
+
+
+current_epoch=0
+model=None
+args = get_arguments()
+
+top1 = AverageMeter()
+top5 = AverageMeter()
+top1.reset()
+top5.reset()
+
+top1_ = AverageMeter()
+top5_ = AverageMeter()
+top1_.reset()
+top5_.reset()
+
+def get_model(args):
+ model = models.resnet50(pretrained=True).eval()
+ model = nn.Sequential(model, nn.Softmax(dim=1))
+ model.cuda()
+ return model
+
+if model is None:
+ model = get_model(args)
+
+
+train_loader, val_loader = data_loader(args, test_path=True)
+
+global_counter = 0
+prob = None
+gt = None
+
+
+y_mask_image = []
+y_image = []
+
+model.cuda()
+
+def denormalize(tensor):
+ means, stds = torch.tensor([0.485, 0.456, 0.406]), torch.tensor([0.229, 0.224, 0.225])
+ denormalized=transforms.Normalize(-1*means/stds,1.0/stds)(tensor)
+ return denormalized
+
+
+target_layer = model[0].layer4[2]
+wrapped_model = GradCAMpp(model, target_layer)
+k = 0
+time_spend = 0
+import matplotlib.image as mpimg 
+
+
+for idx, dat in tqdm(enumerate(val_loader)):
+ img_path, img, label_in = dat
+ im = img
+ global_counter += 1
+ label = label_in
+ img, label = img.cuda(), label.cuda()
+
+ now = time.time()
+ with torch.no_grad():
+ img_n = Metrics.normalize(img)
+ logits = model(img_n)
+ logits0 = logits
+
+ prec1_1, prec5_1 = Metrics.accuracy(logits0.cpu().data, label_in.long(), topk=(1,5))
+ y = logits0.cpu().data.numpy()
+ class_1 = logits0.max(1)[-1] #
+
+ index_gt_y = class_1.long().cpu().data.numpy() #
+
+ Y_i_c = logits0.max(1)[0].item()
+ y_image.append(Y_i_c)
+
+ top1.update(prec1_1[0], img.size()[0])
+ top5.update(prec5_1[0], img.size()[0])
+
+ cam_map,_ = wrapped_model(img_n,index_gt_y)
+ cam_map = F.interpolate(cam_map, size=(224,224), mode='bilinear', align_corners=False)
+ cam_map = Metrics.drop_Npercent(cam_map,args.percentage)
+
+ image = img
+ mask_image = cam_map * image 
+
+ mask_image = mask_image.cuda()
+ with torch.no_grad():
+ img_n = Metrics.normalize(mask_image)
+ logits = model(img_n)
+ logits0 = logits
+
+ prec1_1, prec5_1 = Metrics.accuracy(logits0.cpu().data, label_in.long(), topk=(1,5))
+ y = logits0.cpu().data.numpy()
+
+ Y_i_c_ = logits0[:,index_gt_y][0].item()
+ y_mask_image.append(Y_i_c_)
+
+ top1_.update(prec1_1[0], img.size()[0])
+ top5_.update(prec5_1[0], img.size()[0]) 
+
+y_image = np.array(y_image)
+y_mask_image = np.array(y_mask_image)
+
+
+print('Average Drop is: ', Metrics.AD(y_image,y_mask_image))
+print('Increase in confidence is: ', Metrics.IC(y_image,y_mask_image))
+print('Top1:', top1.avg, 'Top5:',top5.avg)
+print('Top1:', top1_.avg, 'Top5:',top5_.avg)