Broadcast2Pitch

https://github.com/yinmayoo185/soccernetgsr
Admin
This repository contains the official implementation for the SoccerNet Game State Reconstruction...
Tokens:9,643
Snippets:130
Trust Score:5.1
Update:4 months ago
Show doc for...
Tokens:
Raw
### Install RobMOTS Evaluation Code and Run Example

Source: https://github.com/yinmayoo185/soccernetgsr/blob/main/sn-trackeval/docs/RobMOTS-Official/Readme.md

This script installs the TrackEval repository, sets up a virtual environment, installs dependencies, downloads the training ground truth and an example tracker, and runs the evaluation. Requires Python 3.5+.

```bash
# Download the TrackEval repo
git clone https://github.com/JonathonLuiten/TrackEval.git

# Move to repo folder
cd TrackEval

# Create a virtual env in the repo for evaluation
python3 -m venv ./venv

# Activate the virtual env
source venv/bin/activate

# Update pip to have the latest version of packages
pip install --upgrade pip

# Install the required packages
pip install -r requirements.txt

# Download the train gt data
wget https://omnomnom.vision.rwth-aachen.de/data/RobMOTS/train_gt.zip

# Unzip the train gt data you just downloaded.
unzip train_gt.zip

# Download the example tracker 
wget https://omnomnom.vision.rwth-aachen.de/data/RobMOTS/example_tracker.zip

# Unzip the example tracker you just downloaded.
unzip example_tracker.zip

# Run the evaluation on the provided example tracker on the train split (using 4 cores in parallel)
python scripts/run_rob_mots.py --ROBMOTS_SPLIT train --TRACKERS_TO_EVAL STP --USE_PARALLEL True --NUM_PARALLEL_CORES 4

```

--------------------------------

### Setup Project Installation

Source: https://github.com/yinmayoo185/soccernetgsr/blob/main/README.md

Installs the project in 'develop' mode, making it editable and available in the current Python environment.

```bash
python setup.py develop
```

--------------------------------

### Install Project Dependencies

Source: https://github.com/yinmayoo185/soccernetgsr/blob/main/README.md

Installs the main project requirements from the 'requirements.txt' file after installing build dependencies.

```bash
pip install numpy cython
pip install -r requirements.txt
```

--------------------------------

### Install CLIP from Source

Source: https://github.com/yinmayoo185/soccernetgsr/blob/main/README.md

Installs the CLIP library directly from its GitHub repository. This is a specific installation method for the CLIP model.

```bash
pip install git+https://github.com/openai/CLIP.git
```

--------------------------------

### Example Data Line

Source: https://github.com/yinmayoo185/soccernetgsr/blob/main/sn-trackeval/docs/MOTChallenge-Official/Readme.md

An example line demonstrating the 10-value format for object instances in MOT challenge data files.

```text
1, 3, 794.27, 247.59, 71.245, 174.88, -1, -1, -1, -1
1, 6, 1648.1, 119.61, 66.504, 163.24, -1, -1, -1, -1
1, 8, 875.49, 399.98, 95.303, 233.93, -1, -1, -1, -1
...
```

--------------------------------

### Custom Challenge Train/Test Split Example

Source: https://github.com/yinmayoo185/soccernetgsr/blob/main/sn-trackeval/docs/MOTChallenge-Official/Readme.md

Example text files defining train and test splits for a custom challenge by listing sequence names.

```text
name
<seqName1> 
<seqName2>
```

```text
name
<seqName3>
```

--------------------------------

### Custom Challenge Sequence Map Example

Source: https://github.com/yinmayoo185/soccernetgsr/blob/main/sn-trackeval/docs/MOTChallenge-Official/Readme.md

Example of a text file listing sequence names for a custom challenge, used for training splits.

```text
name
<seqName1> 
<seqName2>
<seqName3>
```

--------------------------------

### Install PyTorch with CUDA Support

Source: https://github.com/yinmayoo185/soccernetgsr/blob/main/README.md

Installs PyTorch, torchvision, and torchaudio compatible with CUDA 12.1. Ensure your CUDA version matches the installation command.

```bash
pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
```

--------------------------------

### Example Tracker Result File

Source: https://github.com/yinmayoo185/soccernetgsr/blob/main/sn-trackeval/docs/RobMOTS-Official/Readme.md

A sample of a tracker result file, demonstrating the format for multiple detections across different timesteps.

```text
0 1 3 0.9917707443237305 1200 1920 VaTi0b0lT17F8K3M3N1O1N2O0O2M3N2N101O1O1O01O1O0100O100O01O1O100O10O1000O1000000000000000O1000001O0000000000000000O101O00000000000001O0000010O0110O0O100O1O2N1O2N0O2O2M3M2N2O1O2N5J;DgePZ1
0 2 3 0.989478349685669 1200 1920 Ql^c05ZU12O2N001O0O10OTkNIaT17^kNKaT15^kNLbT14^kNMaT13^kNOaT11_kN0`T10_kN1`T11_kN0`T11_kN0`T1a0O00001O1O1O3M;E5K3M2N000000000O100000000000000000001O00001O2N1O1O1O000001O001O0O2O0O2M3M3M3N2O1O1O1N2O002N1O2N10O02N10000O1O101M3N2N2M7H^_g_1
1 2 3 0.964085042476654 1200 1920 o_Uc03\U12O1O1N102N002N001O1O000O2O1O00002N6J1O001O2N1O3L3N2N4L5K2N1O000000000000001O1O2N01O01O010O01N2O0O2O1M4L3N2N101N2O001O1O100O0100000O1O1O1O2N6I4Mdm^`1
```

--------------------------------

### Tracker Example Code

Source: https://github.com/yinmayoo185/soccernetgsr/blob/main/sn-trackeval/docs/RobMOTS-Official/Readme.md

Example Python code demonstrating how to read in supplied detections and utilize them for tracking. This is useful for creating custom trackers or understanding the integration of supplied detections.

```python
https://github.com/JonathonLuiten/TrackEval/blob/master/trackeval/baselines/stp.py
```

--------------------------------

### Install Deep Person ReID and COCO API

Source: https://github.com/yinmayoo185/soccernetgsr/blob/main/README.md

Installs deep-person-reid and the COCO API from their respective GitHub repositories, using '--no-build-isolation' for specific build requirements.

```bash
pip install git+https://github.com/KaiyangZhou/deep-person-reid.git --no-build-isolation
pip install 'git+https://github.com/cocodataset/cocoapi.git#subdirectory=PythonAPI' --no-build-isolation
```

--------------------------------

### Create Conda Virtual Environment

Source: https://github.com/yinmayoo185/soccernetgsr/blob/main/README.md

Creates a new conda environment named 'SoccernetGSR' with Python 3.12. This is a prerequisite for installing project dependencies.

```bash
conda create -n SoccernetGSR python=3.12
conda activate SoccernetGSR
```

--------------------------------

### Non-Overlap Code Example

Source: https://github.com/yinmayoo185/soccernetgsr/blob/main/sn-trackeval/docs/RobMOTS-Official/Readme.md

Provides access to the Python code for Non-Maximum Suppression (NMS) and the Non-Overlap algorithm used in processing supplied detections. This is recommended for RobMOTS evaluation to ensure tracking results are non-overlapping.

```python
https://github.com/JonathonLuiten/TrackEval/blob/master/trackeval/baselines/non_overlap.py
```

--------------------------------

### Download Model Weights

Source: https://github.com/yinmayoo185/soccernetgsr/blob/main/README.md

Executes a Python script to download necessary pre-trained model weights from Google Drive. Ensure you have internet connectivity.

```bash
python download_properties.py
```

--------------------------------

### Run MOT20 Challenge Evaluation

Source: https://github.com/yinmayoo185/soccernetgsr/blob/main/sn-trackeval/Readme.md

Use this command to evaluate trackers on the MOT20-train benchmark using CLEAR and Identity metrics. Set USE_PARALLEL to False and specify the number of parallel cores.

```bash
python scripts/run_mot_challenge.py --BENCHMARK MOT20 --TRACKERS_TO_EVAL LPC_MOT --METRICS CLEAR Identity --USE_PARALLEL False --NUM_PARALLEL_CORES 1
```

--------------------------------

### Run SoccerNet Game State Evaluation

Source: https://github.com/yinmayoo185/soccernetgsr/blob/main/sn-trackeval/Readme.md

Use this command to perform evaluation for SoccerNet Game State Reconstruction. Ensure the dataset and prediction folders are correctly specified.

```bash
python3 ./scripts/run_soccernet_gs.py --GT_FOLDER path/to/dataset/SoccerNetGS --TRACKERS_FOLDER path/to/predictions_folder --TRACKER_SUB_FOLDER "" --SPLIT_TO_EVAL "test"
```

--------------------------------

### Homography Estimation

Source: https://github.com/yinmayoo185/soccernetgsr/blob/main/README.md

Runs the 'kpts.py' script to generate homography matrices, which are essential for Sports Field Registration. These are saved as .npy files.

```bash
python kpts.py
```

--------------------------------

### Reading MOTS PNG Annotations with PIL

Source: https://github.com/yinmayoo185/soccernetgsr/blob/main/sn-trackeval/docs/MOTS-format.txt

Demonstrates how to read MOTS annotation PNG files using the Pillow library and extract object IDs, class IDs, and instance IDs.

```python
import PIL.Image as Image
import numpy as np

img = np.array(Image.open("000005.png"))
obj_ids = np.unique(img)
# to correctly interpret the id of a single object
obj_id = obj_ids[0]
class_id = obj_id // 1000
obj_instance_id = obj_id % 1000
```

--------------------------------

### Run MOT Challenge Evaluation

Source: https://github.com/yinmayoo185/soccernetgsr/blob/main/sn-trackeval/Readme.md

Use this command to evaluate trackers on the MOT17-train benchmark using CLEAR and Identity metrics. Set USE_PARALLEL to False and specify the number of parallel cores.

```bash
python scripts/run_mot_challenge.py --BENCHMARK MOT17 --TRACKERS_TO_EVAL Lif_T --METRICS CLEAR Identity --USE_PARALLEL False --NUM_PARALLEL_CORES 1
```

--------------------------------

### Run MOT Challenge Evaluation

Source: https://github.com/yinmayoo185/soccernetgsr/blob/main/sn-trackeval/docs/MOTChallenge-Official/Readme.md

Execute the evaluation script for a specified benchmark, split, trackers, and metrics. Supports parallel processing.

```bash
python scripts/run_mot_challenge.py --BENCHMARK MOT17 --SPLIT_TO_EVAL train --TRACKERS_TO_EVAL MPNTrack --METRICS HOTA CLEAR Identity VACE --USE_PARALLEL False --NUM_PARALLEL_CORES 1
```

--------------------------------

### Run Inference for Tracking and Recognition

Source: https://github.com/yinmayoo185/soccernetgsr/blob/main/README.md

Executes the main inference script for player tracking, ReID, and jersey number/role recognition. The 'JERSEY_MODE' (CLIP or LLaMA) should be set in 'configs/config.yaml'.

```bash
python inference_soccernetGSR.py
```

--------------------------------

### Run RobMOTS Evaluation

Source: https://github.com/yinmayoo185/soccernetgsr/blob/main/sn-trackeval/docs/RobMOTS-Official/Readme.md

Execute the evaluation script for your custom tracker on the RobMOTS dataset. Ensure your tracker's results are placed in the correct directory.

```bash
python scripts/run_rob_mots.py --ROBMOTS_SPLIT train --TRACKERS_TO_EVAL YOUR_TRACKER --USE_PARALLEL True --NUM_PARALLEL_CORES 4
```

--------------------------------

### Visualize Prediction Results

Source: https://github.com/yinmayoo185/soccernetgsr/blob/main/README.md

Generates visualizations of the predicted game states based on the output JSON files. Specify the directory containing the JSON results.

```bash
python visualize_prediction_results.py --json_dir results/predicted_SNGamestate_results/test
```

--------------------------------

### Base Metric Class Structure

Source: https://github.com/yinmayoo185/soccernetgsr/blob/main/sn-trackeval/docs/How_To/Add_a_new_metric.md

Illustrates the expected structure for a custom metric class, inheriting from _BaseMetric and defining key functions.

```python
from trackeval.metrics._base_metric import _BaseMetric

class MyCustomMetric(_BaseMetric):

    def __init__(self, **kwargs):
        super().__init__(**kwargs)
        self.fields = [
            'my_custom_field_1', 
            'my_custom_field_2'
        ]
        self.param_keys = ['my_param']

    def eval_sequence(self, data, **kwargs):
        # Calculate metric for a single sequence
        # data is a dictionary containing sequence data
        # returns a dictionary of results
        results = {}
        # ... calculation logic ...
        return results

    def combine_sequences(self, sequences_results, **kwargs):
        # Combine results over sequences
        # sequences_results is a list of results dictionaries from eval_sequence
        # returns a dictionary of combined results
        combined = {}
        # ... combination logic ...
        return combined

    def combine_classes_class_averaged(self, class_results, **kwargs):
        # Combine results over classes (class-averaged)
        # class_results is a dictionary mapping class names to results dictionaries
        # returns a dictionary of combined results
        combined = {}
        # ... combination logic ...
        return combined

    def combine_classes_det_averaged(self, class_results, **kwargs):
        # Combine results over classes (detection-averaged)
        # class_results is a dictionary mapping class names to results dictionaries
        # returns a dictionary of combined results
        combined = {}
        # ... combination logic ...
        return combined

    def _compute_final_fields(self, results, **kwargs):
        # Optional: Compute final fields from intermediate results
        # results is a dictionary of computed fields
        # returns a dictionary of final fields
        final_fields = {}
        # ... final calculation logic ...
        return final_fields
```

--------------------------------

### Evaluate Tracking Results

Source: https://github.com/yinmayoo185/soccernetgsr/blob/main/README.md

Uses the 'sn-trackeval' toolkit to evaluate the prediction results against ground truth data. Ensure the GT_FOLDER and TRACKERS_FOLDER paths are correctly set.

```bash
python ./sn-trackeval/scripts/run_soccernet_gs.py \
    --GT_FOLDER /path/to/SoccerNetGS \
    --TRACKERS_FOLDER /path/to/Pred_SN_GS2025/ \
    --TRACKER_SUB_FOLDER "" \
    --SPLIT_TO_EVAL "test"
```

--------------------------------

### YouTube-VIS Label File Structure

Source: https://github.com/yinmayoo185/soccernetgsr/blob/main/sn-trackeval/docs/YouTube-VIS-format.txt

Defines the JSON structure for label files, including info, videos, annotations, and categories.

```json
{
    "info" : info,
    "videos" : [video],
    "annotations" : [annotation],
    "categories" : [category],
}
```

--------------------------------

### Reading MOTS PNG Annotations with TensorFlow

Source: https://github.com/yinmayoo185/soccernetgsr/blob/main/sn-trackeval/docs/MOTS-format.txt

Shows how to read MOTS annotation PNG files within a TensorFlow input pipeline, decoding them as 16-bit, single-channel images.

```python
import tensorflow as tf

ann_data = tf.read_file(ann_filename)
ann = tf.image.decode_image(ann_data, dtype=tf.uint16, channels=1)
```

--------------------------------

### Load DAVIS Annotation with PIL

Source: https://github.com/yinmayoo185/soccernetgsr/blob/main/sn-trackeval/docs/DAVIS-format.txt

Loads a DAVIS annotation PNG file into a NumPy array using the Pillow library. Ensure the image is treated as a single channel.

```python
import PIL.Image as Image
import numpy as np

img = np.array(Image.open("000005.png"))
```

--------------------------------

### Load DAVIS Annotation with TensorFlow

Source: https://github.com/yinmayoo185/soccernetgsr/blob/main/sn-trackeval/docs/DAVIS-format.txt

Loads and decodes a DAVIS annotation PNG file using TensorFlow, ensuring it's read as a single-channel uint8 image.

```python
import tensorflow as tf

ann_data = tf.read_file(ann_filename)
ann = tf.image.decode_image(ann_data, dtype=tf.uint8, channels=1)
```

--------------------------------

### YouTube-VIS Submission File Structure

Source: https://github.com/yinmayoo185/soccernetgsr/blob/main/sn-trackeval/docs/YouTube-VIS-format.txt

Defines the JSON structure for submission files, containing a list of predictions.

```json
prediction{
    "video_id" : int,
    "category_id" : int,
    "segmentations" : [RLE or [polygon] or None],
    "score" : float,
}
```

--------------------------------

### YouTube-VIS Video Object Structure

Source: https://github.com/yinmayoo185/soccernetgsr/blob/main/sn-trackeval/docs/YouTube-VIS-format.txt

Defines the structure of a video object within the label file.

```json
video{
    "id" : int,
    "width" : int,
    "height" : int,
    "length" : int,
    "file_names" : [file_name],
}
```

--------------------------------

### YouTube-VIS Annotation Object Structure

Source: https://github.com/yinmayoo185/soccernetgsr/blob/main/sn-trackeval/docs/YouTube-VIS-format.txt

Defines the structure of an annotation object within the label file.

```json
annotation{
    "id" : int, 
    "video_id" : int, 
    "category_id" : int, 
    "segmentations" : [RLE or [polygon] or None], 
    "areas" : [float or None], 
    "bboxes" : [[x,y,width,height] or None], 
    "iscrowd" : 0 or 1,
}
```

--------------------------------

### RobMOTS Tracker Result File Format

Source: https://github.com/yinmayoo185/soccernetgsr/blob/main/sn-trackeval/docs/RobMOTS-Official/Readme.md

Format for each row in a tracker result file, representing a single detection. Includes timestep, track ID, class, confidence, image dimensions, and compressed RLE mask.

```text
<Timestep>(int),
<Track ID>(int),
<Class Number>(int),
<Detection Confidence>(float),
<Image Height>(int),
<Image Width>(int),
<Compressed RLE Mask>(string),
```

--------------------------------

### Refine Tracklets using IDATR

Source: https://github.com/yinmayoo185/soccernetgsr/blob/main/README.md

Applies ID-Aware Tracklets Refinement (IDATR) to the initial tracklet results obtained from the inference step.

```bash
./refine_tracklets.sh
```

--------------------------------

### YouTube-VIS Category Object Structure

Source: https://github.com/yinmayoo185/soccernetgsr/blob/main/sn-trackeval/docs/YouTube-VIS-format.txt

Defines the structure of a category object within the label file.

```json
category{
    "id" : int, 
    "name" : str, 
    "supercategory" : str,
}
```

--------------------------------

### MOT Challenge Data Format

Source: https://github.com/yinmayoo185/soccernetgsr/blob/main/sn-trackeval/docs/MOTChallenge-Official/Readme.md

Defines the CSV text-file format for tracker and ground truth data, including frame, ID, bounding box coordinates, and confidence.

```text
<frame>,
<id>,
<bb_left>,
<bb_top>,
<bb_width>,
<bb_height>,
<conf>,
<x>,
<y>,
<z>
```

--------------------------------

### HOTA Metrics Citation

Source: https://github.com/yinmayoo185/soccernetgsr/blob/main/sn-trackeval/docs/RobMOTS-Official/Readme.md

Citation for the HOTA metrics paper. Use this when referencing the HOTA metric in your work.

```bibtex
@article{luiten2020IJCV,
  title={HOTA: A Higher Order Metric for Evaluating Multi-Object Tracking},
  author={Luiten, Jonathon and Osep, Aljosa and Dendorfer, Patrick and Torr, Philip and Geiger, Andreas and Leal-Taix{\'e}, Laura and Leibe, Bastian},
  journal={International Journal of Computer Vision},
  pages={1--31},
  year={2020},
  publisher={Springer}
}
```

--------------------------------

### TrackEval Citation

Source: https://github.com/yinmayoo185/soccernetgsr/blob/main/sn-trackeval/docs/MOTChallenge-Official/Readme.md

BibTeX entry for citing the TrackEval toolkit in academic work.

```bibtex
@misc{luiten2020trackeval,
  author =       {Jonathon Luiten, Arne Hoffhues},
  title =        {TrackEval},
  howpublished = {\url{https://github.com/JonathonLuiten/TrackEval}},
  year =         {2020}
}
```

--------------------------------

### Convert Inference Results to JSON

Source: https://github.com/yinmayoo185/soccernetgsr/blob/main/README.md

Converts the processed inference results into the required JSON format for subsequent evaluation.

```bash
python write_json_file_team.py
```

--------------------------------

### Accessing TrackEval Results

Source: https://github.com/yinmayoo185/soccernetgsr/blob/main/sn-trackeval/docs/RobMOTS-Official/Readme.md

Information on the location and format of tracking evaluation results, including overall summaries and detailed per-sequence data. The 'c', 'd', and 'f' prefixes for metrics denote class averaged, detection averaged, and final geometric mean, respectively.

```text
TrackEval/data/trackers/rob_mots/train/STP/
```

```text
./final_results.csv
```

```text
./results/*
```

--------------------------------

### MOTChallenge Journal Citation

Source: https://github.com/yinmayoo185/soccernetgsr/blob/main/sn-trackeval/docs/MOTChallenge-Official/Readme.md

BibTeX entry for citing the MOTChallenge journal article.

```bibtex
@article{dendorfer2020motchallenge,
  title={MOTChallenge: A Benchmark for Single-camera Multiple Target Tracking},
  author={Dendorfer, Patrick and Osep, Aljosa and Milan, Anton and Schindler, Konrad and Cremers, Daniel and Reid, Ian and Roth, Stefan and Leal-Taix{\'e}, Laura},
  journal={International Journal of Computer Vision},
  pages={1--37},
  year={2020},
}
```
Tokens:
Raw