A neural network model trained with fastai and timm to classify 400 bird species in an image.

This bird classifier is used to predict bird species in a given image. The Image fed should have only one bird. This is a multi-class classification which will output a class even if there is no bird in the image.

Pre-trained model used is Efficient net from timm library, specifically efficientnet_b3a. The dataset trained is from Kaggle BIRDS 400 - SPECIES IMAGE CLASSIFICATION. Evaluation accuracy score on the given test set from Kaggle is 99.4%. Please note this is likely not representative of real world performance, as mentioned by dataset provider that the test set is hand picked as the best images.