This forum is disabled, please visit https://forum.opencv.org
 | 1 | initial version |
As an addition to the answer of mathieu the following comments could be taken into consideration.
The quality of your classifier with 200 positives and 500 negatives depends largely on lots of parameters. - Do the images cover enough lighting conditions? - Do the images cover different scale sizes of the objects to be detected? - How much variation is there in the negative data set? - Did you maintain enough intra-class variance in the training data?
These are all factors defining the final result of your classifier. If dataset is chosen wisely, a training with those numbers could lead to a pretty good classifier/detector of around 75-80% detection rate. However, if you want to improve that, you need to up the amount of training samples drastically.
Usually 200 positives lead to 75% detection rate, but in order to obtain lets say 78% another 1000 positives are needed (yes for only 3% increase) and enough variation should need to be included.
I suggest you read a bit of tutorials online, this will help for sure.
| | 2 | No.2 Revision |
As an addition to the answer of mathieu the following comments could be taken into consideration.
The quality of your classifier with 200 positives and 500 negatives depends largely on lots of parameters.
- parameters.
These are all factors defining the final result of your classifier. If dataset is chosen wisely, a training with those numbers could lead to a pretty good classifier/detector of around 75-80% detection rate. However, if you want to improve that, you need to up the amount of training samples drastically.
Usually 200 positives lead to 75% detection rate, but in order to obtain lets say 78% another 1000 positives are needed (yes for only 3% increase) and enough variation should need to be included.
I suggest you read a bit of tutorials online, this will help for sure.
