Skip to content

Instantly share code, notes, and snippets.

@szrxiao
Last active June 20, 2017 04:18
Show Gist options
  • Save szrxiao/10bd4bfa1acfb752c9b9d8b1beb0845e to your computer and use it in GitHub Desktop.
Save szrxiao/10bd4bfa1acfb752c9b9d8b1beb0845e to your computer and use it in GitHub Desktop.
Data augmentation
Dataset 1: 1048 categories,
1. keep images containing at least one medium/large size logo (>=32*32 pixels)
2. Remove images which aspect ratio >4 or <0.25
3. remove categories with containing less than 20 logos after step 2.
4. random rotate image from step 3 [-10,10] degree 3 times
5. merge the data from step 3 & 4
Results on overall objects (11545)
| IOU | MAP | Th@0.8 | Prec@0.8 | Rec@0.8 | Th@0.9 | Prec@0.9 | Rec@0.9 | Th@0.95 | Prec@0.95 | Rec@0.95 |
| :-- | :----: | :----: | :------: | :-----: | :----: | :------: | :-----: | :-----: | :-------: | :------: |
| 0.3 | 0.6661 | 0.7075 | 0.8001 | 0.6424 | 0.9653 | 0.9001 | 0.4987 | 0.9976 | 0.9502 | 0.2395 |
| 0.4 | 0.6502 | 0.7424 | 0.8 | 0.6265 | 0.9776 | 0.9001 | 0.4612 | 0.9988 | 0.9502 | 0.1619 |
| 0.5 | 0.6172 | 0.8197 | 0.8001 | 0.5882 | 0.9933 | 0.9001 | 0.3449 | 0.9994 | 0.9503 | 0.096 |
Dataset 2: 1048 categories,
1. keep images containing at least one medium/large size logo (>=32*32 pixels)
2. Remove images which aspect ratio >4 or <0.25
3. remove categories with containing less than 20 logos after step 2.
4. random rotate images from step 3 [-10,10] degree 3 times
5. random rotate images from step 4 in 0, 90, 180, or 270 degree
6. merge the data from step 3 & 5
Results on overall objects (11545)
| IOU | MAP | Th@0.8 | Prec@0.8 | Rec@0.8 | Th@0.9 | Prec@0.9 | Rec@0.9 | Th@0.95 | Prec@0.95 | Rec@0.95 |
| :-- | :----: | :----: | :------: | :-----: | :----: | :------: | :-----: | :-----: | :-------: | :------: |
| 0.3 | 0.6699 | 0.7642 | 0.8 | 0.6428 | 0.9705 | 0.9001 | 0.5026 | 0.9986 | 0.9501 | 0.1796 |
| 0.4 | 0.6513 | 0.7976 | 0.8 | 0.6235 | 0.9819 | 0.9 | 0.4577 | 0.9991 | 0.9502 | 0.1304 |
| 0.5 | 0.6203 | 0.8649 | 0.8 | 0.5798 | 0.9955 | 0.9 | 0.308 | 1.0 | 1.0 | 0.0004 |
Dataset 3: 1048 categories,
1. keep images containing at least one medium/large size logo (>=32*32 pixels)
2. Remove images which aspect ratio >4 or <0.25
3. remove categories with containing less than 20 logos after step 2.
4. random rotate images from step 3 [-15,15] degree 3 times
5. random rotate images from step 4 in 0, 90, 180, or 270 degree
6. merge the data from step 3 & 5
| IOU | MAP | Th@0.8 | Prec@0.8 | Rec@0.8 | Th@0.9 | Prec@0.9 | Rec@0.9 | Th@0.95 | Prec@0.95 | Rec@0.95 |
| :-- | :----: | :----: | :------: | :-----: | :----: | :------: | :-----: | :-----: | :-------: | :------: |
| 0.3 | 0.6391 | 0.8537 | 0.8001 | 0.5917 | 0.9821 | 0.9001 | 0.4178 | 0.9986 | 0.9501 | 0.1583 |
| 0.4 | 0.6223 | 0.8759 | 0.8001 | 0.5692 | 0.9888 | 0.9002 | 0.3648 | 0.9991 | 0.9502 | 0.1173 |
| 0.5 | 0.5881 | 0.9147 | 0.8001 | 0.5251 | 0.9966 | 0.9002 | 0.246 | 0.9999 | 0.9595 | 0.0061 |
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment