Merge pull request #12 from LWShowTime/latest_branch

解决重影问题的issue & SOD配置文件调整修改。

Author: NiFangBaAGe

configs/train/segformer/train_segformer_evp_sod.yaml

@@ -58,7 +58,7 @@ optimizer:
   name: adamw
   args:
     lr: 2.e-4
-epoch_max: 20
+epoch_max: 50
 multi_step_lr:
   milestones: [1]
   gamma: 0.1

demo.py

@@ -63,7 +63,7 @@
     img = img.cuda()
 
     pred = model.encoder.forward_dummy(img.unsqueeze(0))
-    pred = torch.sigmoid(pred).view(1, h, w).cpu()
+    pred = torch.sigmoid(pred).view(1, w, h).cpu()
 
     transforms.ToPILImage()(pred).save(args.output)