TFKeras Implementation of Stacked Hourglass Network

A Tensorflow+Keras implementation of Stacked Hourglass Network for Keypoint Detection with several residual layers and can be trained to detect keypoints of chairs and cars.

Dependencies

1. Python 3.6+
2. Tensorflow 2.2.0+ (For GPU, use Tensorflow-GPU 2.2.0+)
3. Pandas

To install these dependencies for CPU, run

pip install -r requirements.txt

(For GPU support, change line tensorflow>=2.2.0 to tensorflow-gpu>=2.2.0 in requirements.txt)

Train Model

To train from scratch, run

python train.py

Note: The train, data, evaluation parameters are present in params.py

Data Format Expected

The dataloader expects a csv file indicating (assuming N keypoints)

<image_name>, <keypointX_1>, <keypointY_1>, <keypointX_2>, <keypointY_2>,.... <keypointX_N>, <keypointY_N>

The image directory (DATA_DIR) and annotation csv (anno_file) need to be specified in params.py.

---

References

1. https://github.com/bearpaw/pytorch-pose

(Visualization is under progress)