How to increase the quality of sampled images ? and does it affect the classifiers accuracy ?

edit

hello, I want to create a object detector .xml file ,for which i gathered some positive images , when i am creating samples -w 20 -h 20 , and when i checked the .vec file the quality of the samples is too bad . I have trained with those images ,and it doesnt detect the object.My concern is that can the quality of sample images be a reason for the failure ? Or the quality of the sample images which are created , downsized and stored in a .vec is always of low quality. The images i used have decent clarity of the object. The positive images I used were downloaded over internet and cropped.

First attempt:- I used 200 unique positive images and 1500 negative images .But the final .xml file does gives false positive. it took around an hour to train with 10 stages. Second attempt:-i used createsample to create 600 sample images from 100 positive images.The result was worse it dosent detect anything. this took only 30 mins to train with 10 stages.

I want to build a Bottle detector , could it be possible that a bottle dosent have so much of unique features to be detected with this algorithm. ?

Also why is the training time so less , i have been reading all over the training time is very long. ?

Thank you