Do you want to try? step-by-step tutorials for beginners

The following tutorials deal with two independent tasks that have different requirements:

• all about training a new CNN model, with a dataset on your own, is highly demanding (i.e. it requires a large amount of calculations). In most cases it requires:
  • a GPU card and all its ecosystem (drivers, CUDA installed,...)
  • a range of software librairies to install on the system (cuDNN, TensorFlow,...)
  • mastering the OS, preferentially Linux
    • using a pre-trained model is much less demanding, does not necessarily requires a GPU card and is much much easier.

In these tutorials, it is possible to skip the training part (when there is one) since the pre-trained models are made available.

NOTA BENE: it should take just the snap of a finger to re-run these analysis on Linux (the widespread OS in the machine learning community). However, some changes/adaptations are required on Windows

Tutorial 1: Detecting giraffe flanks with YOLO (deprecated) (training + detection)

Warning: using Yolo on Windows is highly challenging whereas possible in theory

Tutorial 2: Detecting giraffe flanks with Mask-RCNN (training + detection)

Tutorial 3: Detecting any animal without classification with MegaDetector (detection)

Tutorial 4: Classifying camera trap images with Camera Trap Classifier (classification)

Warning: varying classification performance...

Tutorial 5: Classifying camera trap images with RetinaNet (training + detection/classification)

Tutorial 6: Visual Explanations of a CNN with Grad-CAM (detection/classification + post-treatment)

Tutorial 7: Classifying flower images with Keras (training + classification)