The neural network that receives an image and returns a coarse heat map indicating where the animals are. This NN converts a classification convolutional neural netowork that infers class lables from small image patches to a detection module using 'sliding window' technique. The network is fast to train. Using the provided small dataset, the network is able to finish training under 10 mins with CPU.
If you want to see the network working, run python evaluate.py <image path>
in the network subfolder. For example, python evaluate.py ./dataset_tools/example_raw_data/252.png.
To train the CNN, change the hyperparameters listed in nn_config.yml and execute python train.py
Difficulty Level: Convlutional Neural Network beginners
- opencv
- pytorch
- scipy
- numpy
- matplotlib
- tqdm
- PyYAML
- Pillow