The version 0.4.0 of BirdWatch – the Deep Learning Bird Image Identification System – was released today with several major improvements. The system now supports identification of 312 bird species (see the list of newly added birds below). The new and improved AI model we used in this version also shows significant improvements to the accuracy of the model in the dataset used: with a 98% top-1 accuracy and a 99% top-5 accuracy. This gives you the ability to identify large number of birds more accurately.
The new model was also upgraded to use TensorFlow 2.1, with Python 3.7. We tweaked the data augmentation configuration on the training data to achieve the higer accuracy. You can find more technical details at the The AI Behind BirdWatch page. The accuracy and loss graphs for training and fine-tuning phases of the model is shown beloe:
Following are the new birds added to the system:
| Bird ID | Bird Name |
|---|---|
| 1 | Black-headed Munia |
| 2 | Black-naped Monarch |
| 3 | Black-naped Tern |
| 4 | Black-necked Stork |
| 5 | Black-rumped Flameback |
| 6 | Black-tailed Godwit |
| 7 | Black-throated Munia |
| 8 | Black-winged Kite |
| 9 | Blue Rock Thrush |
| 10 | Blue-eared Kingfisher |
| 11 | Blue-faced Malkoha |
| 12 | Bluethroat |
| 13 | Common Buzzard |
| 14 | Common Coot |
| 15 | Common Greenshank |
| 16 | Common Hawk Cuckoo |
| 17 | Common Hoopoe |
| 18 | Common Iora |
| 19 | Common Kestrel |
| 20 | Common Moorhen |
| 21 | Common Myna |
| 22 | Common Pochard |
| 23 | Common Quail |
| 24 | Common Redshank |
| 25 | Common Ringed Plover |
| 26 | Common Sandpiper |
| 27 | Common Snipe |
| 28 | Common Tailorbird |
| 29 | Common Tern |
| 30 | Common Woodshrike |
| 31 | European Nightjar |
| 32 | Great Bittern |
| 33 | Great Crested Tern |
| 34 | Great Frigatebird |
| 35 | Great Grey Shrike |
| 36 | Great Knot |
| 37 | Great Thick-knee |
| 38 | Great Tit |
| 39 | Great White Pelican |
| 40 | Indian Nightjar |