Skip to content

Commit 9df9c2b

Browse files
Update README.md
1 parent acb8a53 commit 9df9c2b

File tree

1 file changed

+15
-10
lines changed

1 file changed

+15
-10
lines changed

README.md

Lines changed: 15 additions & 10 deletions
Original file line numberDiff line numberDiff line change
@@ -62,18 +62,23 @@ bash deploy.sh
6262

6363
### Result
6464
* Test with method whole and max_size=1600.
65-
* The accuray in paper is hard to reach.
66-
67-
| model | MSE | SAD(normalized by 1000) | link |
68-
| ------------ | ----- | ----------------------- | ---- |
69-
| paper-stage0 | 0.019 | 59.6 | |
70-
| paper-stage1 | 0.017 | 54.6 | |
71-
| paper-stage3 | 0.014 | 50.4 | |
72-
| my-stage0 | 0.035 | 72.9 | [download](https://github.com/huochaitiantang/pytorch-deep-image-matting/releases/download/v1.0/my_stage0_sad_72.9.pth) |
73-
| my-stage1 | 0.033 | 72.2 | |
65+
* The performance in paper is hard to reach.
66+
* training epoch = 25.
67+
* SAD normalized by 1000.
68+
* best epoch is the epoch of the best result while training.
69+
* input img is normalized by mean=[0.485, 0.456, 0.406] and std=[0.229, 0.224, 0.225]
70+
71+
| model | MSE | SAD | best epoch | note | link |
72+
| ------------ | ----- | ---- | ---------- | ---- | ---- |
73+
| paper-stage0 | 0.019 | 59.6 | | | |
74+
| paper-stage1 | 0.017 | 54.6 | | | |
75+
| paper-stage3 | 0.014 | 50.4 | | | |
76+
| my-stage0 | 0.035 | 72.9 | 22 | with crop error and with no normalized input |[download](https://github.com/huochaitiantang/pytorch-deep-image-matting/releases/download/v1.0/my_stage0_sad_72.9.pth) |
77+
| my-stage0 | 0.031 | 70.7 | 19 | with no normalized input | |
78+
| my-stage0 | 0.027 | 69.1 | 12 | fix crop error and with normalized input | |
7479

7580

76-
* visualization, origin image / prediction result / alpha ground truth
81+
* visualization, origin image / prediction result(sad=72.9) / alpha ground truth
7782

7883
* boy-1518482_1920_12.png
7984

0 commit comments

Comments
 (0)