search

Azure / ObjectDetectionUsingCntk

Fast R-CNN Object Detection on Azure using CNTK
128 stars 61 forks source link

Unable to train with template images #28

Closed Somesh58 closed 6 years ago

Somesh58 commented 6 years ago

Hi, I have followed the tutorial and was able to get the desired results for the grocery dataset. However,it works with similar size of the images.I would like to train the model using templates of various sizes of the objects as shown below. 17

I have trained the model using the templates. I have made input images for training(i.e Positive) to a uniform size by padding them with white spaces evenly.Then I have followed the same procedure for labeling and training the images as mentioned in the tutorial.But I am unable to get the results(i.e the model is unable draw detections on the input image).Please suggest a solution for this. I have attached template image, padded image,input image. Template: 17 Templates after Padding: 17 jpg Input Image for detection: 1

Somesh58 commented 6 years ago

@PatrickBue can you please provide a solution for this