deepmac for Animal Masking: Performance on High Noise Images from Animal Camera Traps

In previous posts, it was discussed that Animal Camera Trap data has some data challenges like Illumination, Motion Blur, Small region of interest, Occlusion, Camouflage, Perspective, Weather Conditions, Camera Malfunctions, Temporal Changes, and Non-Animal Variability [1]. These are the images where the model had to resort to putting high feature attribution on the environment as opposed to the animal to produce a prediction. An algorithmic method was proposed to extract such images from a dataset [2]. Then followed up with another post about the performance of Microsoftt’s Object Detection model MegaDetector on such images [3]. MegaDetector was found to be reliable on putting boundary boxes around animals even in noisy images. This post demos the performance of Tensorflow deepmac model for Masking [4]. The image output can be seen below in the Masking Output section. The masking performance is not very reliable in my judgment unless the boundary box is very tight around the animals.

Masking Output

[1] https://arxiv.org/pdf/1807.04975.pdf

[2] https://medium.com/@emad-ezzeldin4/debugging-computer-vision-image-classification-removing-noisy-images-2fb7c518930b

[3] https://medium.com/@emad-ezzeldin4/megadetector-object-detection-performance-on-high-noise-animal-camera-trap-data-30578d23dad3

[4] https://github.com/tensorflow/models/blob/master/research/object_detection/colab_tutorials/deepmac_colab.ipynb