wmcnally / deep-darts

DeepDarts is the first deep learning-based automatic scoring system for steel-tip darts. It predicts dart scores from a single image taken from any camera angle.

DeepDarts

Code for the CVSports 2021 paper: DeepDarts: Modeling Keypoints as Objects for Automatic Scorekeeping in Darts using a Single Camera

Prerequisites

Python 3.5-3.8, CUDA >= 10.1, cuDNN >= 7.6

Setup

1. Install Anaconda or Miniconda
2. Create a new conda environment with Python 3.7: $ conda create -n deep-darts python==3.7. Activate the environment: $ conda activate deep-darts
3. Clone this repo: $ git clone https://github.com/wmcnally/deep-darts.git
4. Go into the directory and install the dependencies: $ cd deep-darts && pip install -r requirements.txt
5. Download images.zip from IEEE Dataport and extract in the dataset directory. Crop the images: $ python crop_images.py --size 800. This step could take a while. Alternatively, you can download the 800x800 cropped images directly from IEEE Dataport. If you choose this option, extract cropped_images.zip in the dataset directory.
6. Download models.zip from IEEE Dataport and extract in the main directory.

Validation / Testing

To test the Dataset 1 model:
$ python predict.py --cfg deepdarts_d1 --split test

To test the Dataset 2 model and write the prediction images:
$ python predict.py --cfg deepdarts_d2 --split test --write

Training

To train the Dataset 1 model:
$ python train.py --cfg deepdarts_d1

To train the Dataset 2 model:
$ python train.py --cfg deepdarts_d2

You may need to adjust the batch sizes to fit your total GPU memory. The default batch sizes are for 24 GB total GPU memory.

Sample Test Predictions

Dataset 1:

Dataset 2:

deep-darts's Issues

How to Detect

Could u explain how to detect with any image?

AttributeError: Can't get attribute 'new_block' on <module 'pandas.core.internals.blocks'

Getting an error running the mentioned train command in the README.md file. Command: python train.py --cfg deepdarts_d1

(deep-darts) C:\Users\phroz\development\deep-darts>python train.py --cfg deepdarts_d1
Running on CPU
Number of accelerators:  1
WARNING:tensorflow:Skipping loading of weights for layer PANetTiny due to mismatch in shape ((1, 1, 512, 30) vs (1, 1, 512, 255)).
WARNING:tensorflow:Skipping loading of weights for layer PANetTiny due to mismatch in shape ((30,) vs (255,)).
WARNING:tensorflow:Skipping loading of weights for layer PANetTiny due to mismatch in shape ((1, 1, 256, 30) vs (1, 1, 256, 255)).
WARNING:tensorflow:Skipping loading of weights for layer PANetTiny due to mismatch in shape ((30,) vs (255,)).
Traceback (most recent call last):
  File "train.py", line 152, in <module>
    yolo = train(cfg, strategy)
  File "train.py", line 99, in train
    batch_size=cfg.train.batch_size * strategy.num_replicas_in_sync)
  File "C:\Users\phroz\development\deep-darts\dataloader.py", line 258, in load_tfds
    data = get_splits(cfg.data.labels_path, cfg.data.dataset, split)
  File "C:\Users\phroz\development\deep-darts\dataloader.py", line 34, in get_splits
    df = pd.read_pickle(path)
  File "C:\Users\phroz\.conda\envs\deep-darts\lib\site-packages\pandas\io\pickle.py", line 187, in read_pickle
    return pc.load(f, encoding=None)
  File "C:\Users\phroz\.conda\envs\deep-darts\lib\site-packages\pandas\compat\pickle_compat.py", line 249, in load
    return up.load()
  File "C:\Users\phroz\.conda\envs\deep-darts\lib\pickle.py", line 1085, in load
    dispatch[key[0]](self)
  File "C:\Users\phroz\.conda\envs\deep-darts\lib\pickle.py", line 1382, in load_stack_global
    self.append(self.find_class(module, name))
  File "C:\Users\phroz\.conda\envs\deep-darts\lib\site-packages\pandas\compat\pickle_compat.py", line 189, in find_class
    return super().find_class(module, name)
  File "C:\Users\phroz\.conda\envs\deep-darts\lib\pickle.py", line 1425, in find_class
    return _getattribute(sys.modules[module], name)[0]
  File "C:\Users\phroz\.conda\envs\deep-darts\lib\pickle.py", line 299, in _getattribute
    .format(name, obj)) from None
AttributeError: Can't get attribute 'new_block' on <module 'pandas.core.internals.blocks' from 'C:\\Users\\phroz\\.conda\\envs\\deep-darts\\lib\\site-packages\\pandas\\core\\internals\\blocks.py'>

Data preparation

Could you tell in more detail about the flow from scratch (data preparation) like the keypoint detection one?

Trained model

Hi there,

is there a chance to get a trained model ?

best

Predict an image

Hi,

How can this be used to predict from an image taken by me?

Kind regards