5,337
edits
Visual Learning and Recognition: Difference between revisions
no edit summary
No edit summary |
|||
| Line 298: | Line 298: | ||
How many octaves? However many octaves to reduces the image size to the template size + 1 for 2x2 upscaling. | How many octaves? However many octaves to reduces the image size to the template size + 1 for 2x2 upscaling. | ||
How many levels? Generally people try 10 levels. | How many levels? Generally people try 10 levels. | ||
===Precision and Recall=== | ===Precision and Recall=== | ||
| Line 364: | Line 354: | ||
The area under the Precision vs Recall curve is call the average precision (AP). | The area under the Precision vs Recall curve is call the average precision (AP). | ||
===Non-max Supression=== | |||
The NMS heuristic here is used to reduce the number of bounding boxes per object to 1. | |||
Initially, you have a set of overlapping bounding boxes <math>B</math>. | |||
Create a final set <math>D</math>. | |||
* While B is not empty | |||
** Remove the highest confidence/score box <math>b_i</math> from <math>B</math>. Add it to <math>D</math> | |||
** For every other box <math>b_j</math>, | |||
*** If <math>IOU(b_i, b_j) > \lambda</math> (i.e. they bound the same object), discard <math>b_j</math> | |||
===Hard mining=== | |||
During training, classify on all images. | During training, classify on all images. | ||
Figure out which instances the classifier classifies incorrectly. | Figure out which instances the classifier classifies incorrectly. | ||
Then train only on those negative instances. | Then train only on those negative instances. | ||
===Current HOG=== | |||
Current HOG uses 31 dimensions | Current HOG uses 31 dimensions | ||
* 9 Contrast insensitive gradients | * 9 Contrast insensitive gradients | ||
* 18 Contrast sensitive gradients | * 18 Contrast sensitive gradients | ||
* 4 | * 4 Texture Related | ||
===Discriminatively Trained Part Based Models (DPM)=== | ===Discriminatively Trained Part Based Models (DPM)=== | ||