diff --git a/CODE_OF_CONDUCT.md b/CODE_OF_CONDUCT.md index f4396c3..339757e 100644 --- a/CODE_OF_CONDUCT.md +++ b/CODE_OF_CONDUCT.md @@ -55,7 +55,7 @@ further defined and clarified by project maintainers. ## Enforcement Instances of abusive, harassing, or otherwise unacceptable behavior may be -reported by contacting the project team at alexander.mathis@bethgelab.org. All +reported by contacting the project team at alexander.mathis@epfl.ch. All complaints will be reviewed and investigated and will result in a response that is deemed necessary and appropriate to the circumstances. The project team is obligated to maintain confidentiality with regard to the reporter of an incident. diff --git a/README.md b/README.md index 138c03b..246783a 100644 --- a/README.md +++ b/README.md @@ -47,6 +47,16 @@ code: https://github.com/PolarBean/DLC_ROI_tool Contributed by [Harry Carey](https://github.com/PolarBean) +## Linear Transformation and Scaling of DLC output data (transform_and_scale) + +This package is designed for anyone who wants to know where a tracked marker is within a reference frame (i.e. behavioral context). DeepLabCut outputs coordinates in relation to the field of view of the recorded video. With this tool, these coordinates can be linearly transformed and scaled to the reference frame of the behavioral context, meaning that the output coordinates are distances [cm] to a corner of the behavioral context, instead of distances [px] to a corner of the video field of view. + +code: https://github.com/DeepLabCut/DLCutils/tree/master/transform_and_scale/ + +tutorial: https://github.com/DeepLabCut/DLCutils/tree/master/transform_and_scale/transform_and_scale_tutorial.ipynb + +Contributed by [Michael Schellenberger](https://github.com/MSchellenberger) + # Clustering tools (using the output of DLC): ## Identifying Behavioral Structure from Deep Variational Embeddings of Animal Motion diff --git a/transform_and_scale/DLCTransformer.py b/transform_and_scale/DLCTransformer.py new file mode 100644 index 0000000..b53226a --- /dev/null +++ b/transform_and_scale/DLCTransformer.py @@ -0,0 +1,208 @@ +import pandas as pd +import numpy as np +from read_config import read_config +from typing import Optional + +class DLCTransformer: + def __init__(self, config_filepath: str, dlc_filepath: Optional[str] = None, dlc_df: Optional[pd.DataFrame] = None): + self.config = read_config(config_filepath) + self.dlc_filepath = dlc_filepath + if self.dlc_filepath is not None: + if self.dlc_filepath.endswith("h5"): + self.dlc_df = pd.read_hdf(dlc_filepath, header=[0, 1, 2], index_col=0) + elif self.dlc_filepath.endswith("csv"): + self.dlc_df = pd.read_csv(dlc_filepath, header=[0, 1, 2], index_col=0) + else: + raise ValueError("DeepLabCut file must be .h5 or .csv") + else: + if dlc_df is not None: + self.dlc_df = dlc_df + else: + raise ValueError("One of the arguments dlc_filepath or dlc_df must be specified") + + # read metadata + self.origin_marker = self.config["origin_marker"] + self.basis_vector_h_marker = self.config["basis_vector_h_marker"] + self.basis_vector_v_marker = self.config["basis_vector_v_marker"] + self.scale_factor_h = self.config["scale_factor_h"] + self.scale_factor_v = self.config["scale_factor_v"] + self.dlc_immobile_marker_threshold = self.config[ + "dlc_immobile_marker_threshold" + ] + self.show_angle = self.config["show_angle"] + self.scorer = self.dlc_df.columns.get_level_values(0).unique()[0] + + def run(self) -> pd.DataFrame: + """ + Translate and scale DLC data. + First get median coordinate of origin marker and basis vectors, then transform and scale the data. + + :return: pd.DataFrame with transformed and scaled DLC data + """ + # get basis vectors and origin + origin, basis_vector_h, basis_vector_v = self.get_basis_vectors( + df=self.dlc_df, + origin_marker=self.origin_marker, + basis_vector_h_marker=self.basis_vector_h_marker, + basis_vector_v_marker=self.basis_vector_v_marker, + dlc_immobile_marker_threshold=self.dlc_immobile_marker_threshold, + ) + + if self.show_angle: + print( + "Angle between basis vectors :", + self._calculate_angle(basis_vector_v, (0, 0), basis_vector_h), + ) + + # transform df + df_transformed = self.transform( + self.dlc_df, origin, basis_vector_h, basis_vector_v + ) + df_scaled = self.scale_df( + df_transformed, self.scale_factor_h, self.scale_factor_v + ) + return df_scaled + + def get_basis_vectors( + self, + df: pd.DataFrame, + origin_marker: str, + basis_vector_h_marker: str, + basis_vector_v_marker: str, + dlc_immobile_marker_threshold: float, + ) -> (tuple, tuple, tuple): + """ + Returns the origin and basis vectors of the coordinate system + + :param df: pd.DataFrame with tracking data + :param origin_marker: name of origin marker + :param basis_vector_h_marker: name of horizontal basis vector marker + :param basis_vector_v_marker: name of vertical basis vector marker + :param dlc_immobile_marker_threshold: minimum likelihood of a constant marker to be included into the median coordinate calculation + :return: coordinates of origin, basis_vector_h, basis_vector_v + """ + origin = self.get_median_coordinate( + df, origin_marker, dlc_immobile_marker_threshold + ) + + # bring coordinates into origin system + basis_vector_h_coord = self.get_median_coordinate( + df=df, + marker=basis_vector_h_marker, + dlc_immobile_marker_threshold=dlc_immobile_marker_threshold, + ) + basis_vector_v_coord = self.get_median_coordinate( + df=df, + marker=basis_vector_v_marker, + dlc_immobile_marker_threshold=dlc_immobile_marker_threshold, + ) + + # calculate basis vector transformation + basis_vector_h = ( + basis_vector_h_coord[0] - origin[0], + basis_vector_h_coord[1] - origin[1], + ) + basis_vector_v = ( + basis_vector_v_coord[0] - origin[0], + basis_vector_v_coord[1] - origin[1], + ) + + return origin, basis_vector_h, basis_vector_v + + def get_median_coordinate(self, df, marker, dlc_immobile_marker_threshold) -> tuple: + """ + Returns the most likely coordinate of a vector + + :param df: pd.DataFrame with tracking data + :param marker: name of marker of which the median coordinate should be calculated + :param dlc_immobile_marker_threshold: minimum likelihood of a constant marker to be included into the median coordinate calculation + :return: median coordinate of marker + """ + # filter df + df = df.droplevel(0, axis=1) + marker_df = df.loc[ + df[marker, "likelihood"] > dlc_immobile_marker_threshold, marker + ].copy() + x_coord = np.nanmedian(marker_df["x"]) + y_coord = np.nanmedian(marker_df["y"]) + coords = (x_coord, y_coord) + return coords + + def _calculate_angle(self, a, b, c) -> float: + """ + Calculates the angle between three points a-b-c + + :param a: tuple of coordinates + :param b: tuple of coordinates + :param c: tuple of coordinates + :return: angle between a-b-c in degrees + """ + + # Calculate the vectors AB and BC + vector_AB = (b[0] - a[0], b[1] - a[1]) + vector_BC = (c[0] - b[0], c[1] - b[1]) + + # Calculate the magnitudes of AB and BC + magnitude_AB = np.linalg.norm(vector_AB) + magnitude_BC = np.linalg.norm(vector_BC) + # + # Calculate the dot product of AB and BC + dot_product = np.dot(vector_AB, vector_BC) + + # Calculate the angle in radians using the dot product and magnitudes + angle_radians = np.arccos(dot_product / (magnitude_AB * magnitude_BC)) + + # Convert the angle to degrees + angle_degrees = np.degrees(angle_radians) + return angle_degrees + + def transform(self, df, origin, basis_vector_h, basis_vector_v) -> pd.DataFrame: + """ + Transform the coordinates of the df into the new coordinate system + + :param df: pd.DataFrame with tracking data + :param origin: coordinate of origin + :param basis_vector_h: coordinate of horizontal basis vector + :param basis_vector_v: coordinate of vertical basis vector + :return: transformed pd.DataFrame + """ + transformed_df = df.copy() + final_transformed_df = df.copy() + + # 2d rotation matrix + a = basis_vector_h[0] + b = basis_vector_h[1] + c = basis_vector_v[0] + d = basis_vector_v[1] + + for marker in df.columns.get_level_values(1).unique(): + + # shift coordinates into origin system + transformed_df.loc[:, (self.scorer, marker, "x")] -= origin[0] + transformed_df.loc[:, (self.scorer, marker, "y")] -= origin[1] + + # linear algebra 2D transformation + final_transformed_df.loc[:, (self.scorer, marker, "y")] = ( + (transformed_df.loc[:, (self.scorer, marker, "y")] / b) + - (transformed_df.loc[:, (self.scorer, marker, "x")] / a) + ) / ((-c / a) + (d / b)) + + final_transformed_df.loc[:, (self.scorer, marker, "x")] = ( + transformed_df.loc[:, (self.scorer, marker, "x")] + - c * final_transformed_df.loc[:, (self.scorer, marker, "y")] + ) / a + + return final_transformed_df + + def scale_df(self, df, scale_factor_h, scale_factor_v) -> pd.DataFrame: + """ + :param df: pd.DataFrame with tracking data + :param scale_factor_h: horizontal scale factor + :param scale_factor_v: vertical scale factor + :return: scaled pd.DataFrame + """ + scaled_df = df.copy() + for marker in df.columns.get_level_values(1).unique(): + scaled_df.loc[:, (self.scorer, marker, "x")] *= scale_factor_h + scaled_df.loc[:, (self.scorer, marker, "y")] *= scale_factor_v + return scaled_df diff --git a/transform_and_scale/README.md b/transform_and_scale/README.md new file mode 100644 index 0000000..fae5edc --- /dev/null +++ b/transform_and_scale/README.md @@ -0,0 +1,28 @@ +# Easy transformation and scaling of DeepLabCut Data + +#### Background +DeepLabCut is a widely used markerless pose estimation toolbox in behavioral science. The output of DeepLabCut is coordinates in pixels for each frame for each marker. However, in most cases the coordinates first need to be translated (i.e. adapted to the coordinate-space of the behavioral maze) and scaled (e.g. to cm) to enable meaningful behavioral quantification. + +#### Functionality +The Transform_DLC repository takes care of a specific task: It takes DeepLabCut dataframes as the input and outputs transformed and scaled DeepLabCut dataframes. Requirements for this transformation and scaling are that the behavioral maze or a reference has to be rectangular and its corners have to be tracked with DeepLabCut. + +#### Usage +For an example usecase check out the tutorial notebook! + +1) Set hyperparameter in config file: Specify the name of the horizontal and vertical basis vector markers (in a rectangular maze these are two opposing corners) and the origin (the corner that connects the two basis vector corners). Additionally, set the DeepLabCut-likelihood threshold for those markers, and the scale factors (optional, if you don´t need scaling set them to 1). For quality control set show_angle to True. + +2) Instantiate the DLCTransformer class with the filepath to your DeepLabCut tracked data and the config-filepath. + +3) Use .run on your instantiated class object and save the output in a variable of your choice. + + +#### Contribution +This is a [Defense Circuits Lab](https://www.defense-circuits-lab.com/) project written by [Michael Schellenberger](https://github.com/MSchellenberger) for [DLCutils](https://github.com/DeepLabCut/DLCutils). + + + +
+ + DefenseCircuitsLab + +
\ No newline at end of file diff --git a/transform_and_scale/__pycache__/DLCTransformer.cpython-39.pyc b/transform_and_scale/__pycache__/DLCTransformer.cpython-39.pyc new file mode 100644 index 0000000..8f81d28 Binary files /dev/null and b/transform_and_scale/__pycache__/DLCTransformer.cpython-39.pyc differ diff --git a/transform_and_scale/__pycache__/read_config.cpython-39.pyc b/transform_and_scale/__pycache__/read_config.cpython-39.pyc new file mode 100644 index 0000000..8f303f8 Binary files /dev/null and b/transform_and_scale/__pycache__/read_config.cpython-39.pyc differ diff --git a/transform_and_scale/config.yaml b/transform_and_scale/config.yaml new file mode 100644 index 0000000..763d295 --- /dev/null +++ b/transform_and_scale/config.yaml @@ -0,0 +1,7 @@ +origin_marker: 'RodLowerLeft' +basis_vector_h_marker: 'RodLowerRight' +basis_vector_v_marker: 'RodUpperLeft' +scale_factor_h: 5.8 +scale_factor_v: 3 +dlc_immobile_marker_threshold: 0.9 +show_angle: True \ No newline at end of file diff --git a/transform_and_scale/read_config.py b/transform_and_scale/read_config.py new file mode 100644 index 0000000..94b43e6 --- /dev/null +++ b/transform_and_scale/read_config.py @@ -0,0 +1,21 @@ +import yaml + + +def read_config(config_path: str) -> dict: + """ + Reads structured config file defining a project. + + :param config_path: path to config file + """ + + try: + with open(config_path, "r") as ymlfile: + config_file = yaml.load(ymlfile, Loader=yaml.SafeLoader) + except FileNotFoundError: + raise ( + "Could not find the config file at " + + config_path + + " \n Please make sure the path is correct and the file exists" + ) + + return config_file diff --git a/transform_and_scale/test_data/DefenseCircuitsLab2023_Rotarod_testfile.h5 b/transform_and_scale/test_data/DefenseCircuitsLab2023_Rotarod_testfile.h5 new file mode 100644 index 0000000..f4c6184 Binary files /dev/null and b/transform_and_scale/test_data/DefenseCircuitsLab2023_Rotarod_testfile.h5 differ diff --git a/transform_and_scale/transform_and_scale_tutorial.ipynb b/transform_and_scale/transform_and_scale_tutorial.ipynb new file mode 100644 index 0000000..a2b9ed3 --- /dev/null +++ b/transform_and_scale/transform_and_scale_tutorial.ipynb @@ -0,0 +1,63 @@ +{ + "cells": [ + { + "cell_type": "code", + "execution_count": null, + "metadata": {}, + "outputs": [], + "source": [ + "from DLCTransformer import DLCTransformer" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "metadata": {}, + "outputs": [], + "source": [ + "dlc_filepath = r\"test_data/DefenseCircuitsLab2023_Rotarod_testfile.h5\"\n", + "config_filepath = r\"config.yaml\"\n", + "transformer = DLCTransformer(dlc_filepath=dlc_filepath, config_filepath=config_filepath)" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "metadata": {}, + "outputs": [], + "source": [ + "scaled = transformer.run()" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "metadata": {}, + "outputs": [], + "source": [ + "scaled.head()" + ] + } + ], + "metadata": { + "kernelspec": { + "display_name": "Python 3 (ipykernel)", + "language": "python", + "name": "python3" + }, + "language_info": { + "codemirror_mode": { + "name": "ipython", + "version": 3 + }, + "file_extension": ".py", + "mimetype": "text/x-python", + "name": "python", + "nbconvert_exporter": "python", + "pygments_lexer": "ipython3", + "version": "3.9.12" + } + }, + "nbformat": 4, + "nbformat_minor": 1 +}