Closed khyatibhuva closed 3 years ago
Challenges and solutions:
The first challenge was to prepare dataset. The dataset we found was not in class-wise format and to train our model we required data into classwise format. So, first we have to prepare the data class-wise. Also, the images were not of the same size and in most of the images background was there. So, we also have to make all images of same size as well as we have to remove background. To solve this issue first i created dataset classwise using file handling techniques and with the help of opencv i made the all images of same size and removed the background from all images.
While training the model we come to know that data is not enough to train the model. For training such type of model , we require the big dataset of images to classify classes well. To solve this issue i have done data augmentation in our data. From our original data i have generated more images of same type as well as i have done some reshuffling in training and testing data.
After training model the issue we faced was, model was correctly working and giving good accuracy in our system but when the integration team were trying predict the images the model was giving wrong output each and everytime. The main problem was with the ordering of classes in both the system. Both the system were taking different order of the classes. After making ordering same this isuue was resolved.
Nicely done.
Write your challenges you faced in this entire project and how to solved them.