Detection of drones using FastRCNN via CNTK's implemententation

(Install Anaconda previously or let the install-cntk.sh do it for you, recommended to do it previously)

https://www.anaconda.com/download/#linux

Download CNTK-2.3.1 (CPU only) tar.gz = https://cntk.ai/dllc-2.3.html Extract it out.

cd /home/username/cntk/Scripts/install/linux

./install-cntk.sh --py-version 34

Above command will make an env called cntk-py34

Activate above env by command: source activate cntk-py34

Activate CNTK env by command: source "/home/slapbot/cntk/activate-cntk"

Clone the repo: git clone https://github.com/SlapBot/Drone-Detectron.git

Cd in: cd drone-detection/Detection/FastRCNN

Install the AlexNet Model: python install_fastrcnn.py

Install its deps: pip install -r requirements.txt

Install Scikit-Image: pip install scikit-image

Remove pre-installed opencv from conda (bug) and install latest one from pip via commands:

conda remove opencv

pip install opencv-python

Run Selective Search by command: python A1_GenerateInputROIs.py

Visualize the selective search results: python B1_VisualizeInputROIs.py

Check recall of proposed regions found in selective search task: python B2_EvaluateInputROIs.py

Finally train the model using:

cntk configFile=/home/username/drone-detection/Detection/FastRCNN/proc/Drones_500/cntkFiles/fastrcnn.cntk currentDirectory=/home/username/drone-detection/Detection/FastRCNN/proc/Drones_500/cntkFiles/ NumLabels=3 NumTrainROIs=500 TrainROIDim=2000 TrainROILabelDim=1500 NumTestROIs=500 TestROIDim=2000 TestROILabelDim=1500

Evaluate and See the results using: python evaluateDetections.py

Change the value count to any value upto which you wanna see the results at:

drone-detection/Detection/FastRCNN/imdb_data.py L:236 visualize_multiple(visualizers[2:], count=10)

https://github.com/creiser/drone-detection

Used creiser's original repository and made various dependency and code changes to get it back to working.