Embedded Files
Few key challenges faced by us were
Few key challenges faced by us were
If we tested our model on a completely new kind of flower or object which had different texture and style as compared to training set, then our model performed poorly.
This can be mitigated by training the models on larger datasets for longer times.
We observed that yellow and brown were the predominant colors in our training set. Therefore, our model used these colors heavily in our test results. That being said, the results were very appealing.
Inconsistent coloring scheme was observed for dense or uncertain texture
Images from Training set
Images from Test set
If the network size was small then it could not generalize well. We faced the problem of overfitting on the training set.
Lack of Computational Resources
Lack of Computational Resources
Training a deep neural network on a huge amount of data requires substantial amount of time and resources. State of the art models have been trained on Image Net dataset for hundreds of hours.
Due to the limited resources, we trained our models on a substantially smaller dataset (1500 images) for 100 epochs each. We used Google Colab Pro for training our models.
Page updated
Google Sites
Report abuse