Could you provide some example code that you used to train the network (i.e. usage of train.py and test.py)? And to confirm, you are applying Faster-RCNN from the MXNet examples, to classes from the VOC dataset (pedestrians and bicycle)? Since no bicycles are found the average precision cannot be calculated, hence the error you’re seeing. Giving the network more training epochs may help here.
Thank you for your reply… I am using the same example code.
When i train the network, i’m not gettign the good model for testing on some demo examples. It is detecting both objects as only pedestrians.
I am using the saem train.py and test.py adn demo.py scripts…
I just changed class names to ‘pedestrian’, ‘bicycle’ along with background class.
i have created dataset of 800 images of each class…my data is greyscale ( channel = 1)
My data is from radar… so range doppler maps are my data images… each image will have some blobs…
Before i was able to train properly but testing was not happening…
but now training itself is not proper…
Can you suggest me the solution for this…
Please provide me your mail id if you need any details, I will give more inputs from my side… Since it is very confidential thing…