Pedestrian Detection using the TensorFlow Object Detection API and Nanonets. [Blog][Performance]
This repo provides complementary material to this blog post, which compares the performance of four object detectors for a pedestrian detection task. It also introduces a feature to use multiple GPUs in parallel for inference using the multiprocessing package. The count accuracy and FPS for different models (using 1,2,4 or 8 GPUs in parallel) were calculated and plotted.
The TownCentre dataset is used for training our pedestrian detector. You can use the following commands to download the dataset. This automatically extracts the frames from the video, and creates XML files from the csv groundtruth. The image dimensions are downscaled by a factor of 2 to reduce processing overhead.
wget http://www.robots.ox.ac.uk/ActiveVision/Research/Projects/2009bbenfold_headpose/Datasets/TownCentreXVID.avi
wget http://www.robots.ox.ac.uk/ActiveVision/Research/Projects/2009bbenfold_headpose/Datasets/TownCentre-groundtruth.top
python extract_towncentre.py
python extract_GT.py
Refer to the instructions in this blog post.
Step 1: Clone the repo
git clone https://github.com/NanoNets/object-detection-sample-python.git
cd object-detection-sample-python
sudo pip install requests
Step 2: Get your free API Key
Get your free API Key from http://app.nanonets.com/user/api_key
Step 3: Set the API key as an Environment Variable
export NANONETS_API_KEY=YOUR_API_KEY_GOES_HERE
Step 4: Create a New Model
python ./code/create-model.py
Note: An environment variable NANONETS_MODEL_ID will be created in the previous step, with your model ID.
Step 5: Upload the Training Data
Place the training data in a folder named images and annotations in annotations/json
python ./code/upload-training.py
Step 6: Train the Model
python ./code/train-model.py
Step 7: Get Model State
The model takes ~2 hours to train. You will get an email once the model is trained. In the meanwhile you check the state of the model
python ./code/model-state.py
Step 8: Make Predictions
Create a folder named 'test_images' inside the 'nanonets' folder. Place the input images in this folder, and then run this command.
python ./code/prediction.py
For more stats, refer to the blog post. The performance of each model (on the test set) was compiled into a video, which you can see here.
In light of GDPR and feeble accountability of Deep Learning, it is imperative that we ponder about the legality and ethical issues concerning automation of surveillance. This blog/code is for educational purposes only, and it used a publicly available dataset. It is your responsibility to make sure that your automated system complies with the law in your region.