multiple-digit-recognizer

Here i am using sliding window technic to select all the possible part of image and checking for actual image vs background image. Then if it is not a background image then passing it to another model which detect its value(digit no. 0 to 9).After that using IOU and NON MAX SUPPRESION eleminate all unwanted predictions we have our final output image. And i have got about 98 % accuracy

Input image

First we have to change our input image according to our training image(MNIST DATASET)

After passing this image to our model we got this image

Now we apply NMS to our image then got the final output image