first run

opt.py

 parse = argparse.ArgumentParser(description='PyTorch Polyp Segmentation')
 
 "-------------------data option--------------------------"
-parse.add_argument('--root', type=str, default='/scratch/krushi1992/MICCAI2021/new-med-up/CVC-frame/')
-parse.add_argument('--dataset', type=str, default='EndoScene')
-parse.add_argument('--train_data_dir', type=str, default='train')
-parse.add_argument('--valid_data_dir', type=str, default='valid')
-parse.add_argument('--test_data_dir', type=str, default='test')
+# parse.add_argument('--root', type=str, default='/media/external_10TB/10TB/pourmand/CVC-EndoSceneStill')
+# parse.add_argument('--dataset', type=str, default='EndoScene')
 
+parse.add_argument('--root',default='/media/external_10TB/10TB/pourmand/Kvasir-SEG/Kvasir-SEG')
+parse.add_argument('--dataset',type=str,default='kvasir_SEG')
 
+parse.add_argument('--train_data_dir', type=str, default='')
+parse.add_argument('--valid_data_dir', type=str, default='')
+parse.add_argument('--test_data_dir', type=str, default='')
 
 "-------------------training option-----------------------"
 parse.add_argument('--mode', type=str, default='train')
-parse.add_argument('--nEpoch', type=int, default=200)
+parse.add_argument('--nEpoch', type=int, default=10)
 parse.add_argument('--batch_size', type=float, default=4)
 parse.add_argument('--num_workers', type=int, default=0)
 parse.add_argument('--use_gpu', type=bool, default=True)
 parse.add_argument('--load_ckpt', type=str, default=None)
 parse.add_argument('--model', type=str, default='EUNet')
 parse.add_argument('--expID', type=int, default=1)
-parse.add_argument('--ckpt_period', type=int, default=50)
+parse.add_argument('--ckpt_period', type=int, default=0)
 parse.add_argument('--weight_const', type=float, default=0.3)
 
 "-------------------optimizer option-----------------------"

results.txt

+recall: 0.8576, specificity: 0.9496, precision: 0.7379, F1: 0.7466, F2: 0.7929, ACC_overall: 0.9293, IoU_poly: 0.6324, IoU_bg: 0.9147, IoU_mean: 0.7736
+recall: 0.9194, specificity: 0.9545, precision: 0.7454, F1: 0.7950, F2: 0.8533, ACC_overall: 0.9426, IoU_poly: 0.6884, IoU_bg: 0.9293, IoU_mean: 0.8089
+recall: 0.9209, specificity: 0.9635, precision: 0.8065, F1: 0.8338, F2: 0.8733, ACC_overall: 0.9514, IoU_poly: 0.7441, IoU_bg: 0.9394, IoU_mean: 0.8417
+recall: 0.9211, specificity: 0.9660, precision: 0.8323, F1: 0.8523, F2: 0.8830, ACC_overall: 0.9552, IoU_poly: 0.7704, IoU_bg: 0.9440, IoU_mean: 0.8572
+recall: 0.8685, specificity: 0.9803, precision: 0.8929, F1: 0.8590, F2: 0.8577, ACC_overall: 0.9586, IoU_poly: 0.7808, IoU_bg: 0.9483, IoU_mean: 0.8645
+recall: 0.9076, specificity: 0.9653, precision: 0.8430, F1: 0.8484, F2: 0.8723, ACC_overall: 0.9523, IoU_poly: 0.7675, IoU_bg: 0.9407, IoU_mean: 0.8541
+recall: 0.9342, specificity: 0.9678, precision: 0.8530, F1: 0.8732, F2: 0.9004, ACC_overall: 0.9605, IoU_poly: 0.8002, IoU_bg: 0.9497, IoU_mean: 0.8750
+recall: 0.9368, specificity: 0.9719, precision: 0.8748, F1: 0.8886, F2: 0.9102, ACC_overall: 0.9658, IoU_poly: 0.8212, IoU_bg: 0.9558, IoU_mean: 0.8885
+recall: 0.9421, specificity: 0.9756, precision: 0.8770, F1: 0.8940, F2: 0.9169, ACC_overall: 0.9680, IoU_poly: 0.8276, IoU_bg: 0.9585, IoU_mean: 0.8930
+recall: 0.9344, specificity: 0.9764, precision: 0.8932, F1: 0.8986, F2: 0.9143, ACC_overall: 0.9691, IoU_poly: 0.8364, IoU_bg: 0.9599, IoU_mean: 0.8982

train.py

 
 
 def train():
+    file_name = 'results.txt'
 
     model = generate_model(opt)
     #model = nn.DataParallel(model)
                  metrics_result['F1'], metrics_result['F2'], metrics_result['ACC_overall'],
                  metrics_result['IoU_poly'], metrics_result['IoU_bg'], metrics_result['IoU_mean']))
 
+        with open(file_name,'a') as f:
+            f.write('recall: %.4f, specificity: %.4f, precision: %.4f, F1: %.4f,'
+              ' F2: %.4f, ACC_overall: %.4f, IoU_poly: %.4f, IoU_bg: %.4f, IoU_mean: %.4f'
+              % (metrics_result['recall'], metrics_result['specificity'], metrics_result['precision'],
+                 metrics_result['F1'], metrics_result['F2'], metrics_result['ACC_overall'],
+                 metrics_result['IoU_poly'], metrics_result['IoU_bg'], metrics_result['IoU_mean'])+'\n')
+
+
         if ((epoch + 1) % opt.ckpt_period == 0): 
             torch.save(model.state_dict(), './checkpoints/exp' + str(opt.expID)+"/ck_{}.pth".format(epoch + 1))
 

utils/__init__.py

+from .transform import *

utils/loss.py

     criterion = BceDiceLoss()
 
     loss0 = criterion(d0, gt)
-    gt = F.interpolate(gt, scale_factor=0.5, mode='bilinear', align_corners=True, recompute_scale_factor=True)
+    gt = F.interpolate(gt, scale_factor=0.5, mode='bilinear', align_corners=True)
     loss1 = criterion(d1, gt)
-    gt = F.interpolate(gt, scale_factor=0.5, mode='bilinear', align_corners=True, recompute_scale_factor=True)
+    gt = F.interpolate(gt, scale_factor=0.5, mode='bilinear', align_corners=True)
     loss2 = criterion(d2, gt)
-    gt = F.interpolate(gt, scale_factor=0.5, mode='bilinear', align_corners=True, recompute_scale_factor=True)
+    gt = F.interpolate(gt, scale_factor=0.5, mode='bilinear', align_corners=True)
     loss3 = criterion(d3, gt)
-    gt = F.interpolate(gt, scale_factor=0.5, mode='bilinear', align_corners=True, recompute_scale_factor=True)
+    gt = F.interpolate(gt, scale_factor=0.5, mode='bilinear', align_corners=True)
     loss4 = criterion(d4, gt)
 
     return loss0 + loss1 + loss2 + loss3 + loss4