Mothe is a python-based application with GUI to detect and track multiple animals in heterogeneous environments. It has been tested on Linux/Windows/Mac OS.
Reference: Rathore A, Sharma A, Shah S, Sharma N, Torney C, Guttal V. 2023. Multi-Object Tracking in Heterogeneous environments (MOTHe) for animal video recordings. PeerJ 11:e15573, doi: https://peerj.com/articles/15573/ (also at bioRxiv: https://www.biorxiv.org/content/10.1101/2020.01.10.899989v2.full)
MOTHe uses Convolutional Neural Network (CNN) architecture for the object detection task. It takes a digital image as an input and reads its features to assign a category. Learning algorithms extract features from the images by using huge amounts of labeled training data. Once we train the neural network to generate models of the objects of interest, they can be used to classify novel data (images). MOTHe is designed to be generic which empowers the user to track objects of interest even in natural settings.
Please check all the pre-requisites and installation steps before processding to MOTHe-GUI demonstration. While we provide detailed instructions, basic familiarity with command prompts in terminal (in any OS) will be helpful.
You need to install anaconda and ensure all dependneices are installed. Please visit this wiki page to learn about installing anaconda for your respective OS. Follow the link below.
https://github.com/tee-lab/MOTHe-GUI/wiki/MOTHe-GUI-wiki
Please make sure to read above section even if you have anaconda already installed in your system, as there are many dependencies that may require installations.
-
Download/clone this repository to your computer.
-
For the purpose of trial/learning of our package, you can run MOTHe on our sample videos. Download and copy untracked videos from here to the same folder in which you have kept the MOTHe package (MOTHe-GUI-master). To begin with, you can download only one of the videos provided (wasp or blackbuck).
-
Enter the anaconda environment:
Windows OS: Start the Anaconda shell from the start menu.
Linux/Mac OS: o\Open the terminal, then type:
$ conda activate mothe
screenshots for win/linux required
- Change the directory to MOTHe folder:
$ cd <path to MOTHe-GUI directory>
- Run MOTHe GUI app:
$ python mothe_gui.py
It might take some time (few seconds to a minute) to launch the GUI as it uses the tensorflow library.
Follow the detailed steps given below to run different functions of the app
Step 1: Configuration
This step is used to set parameters of MOTHe. Parameters to be set in this step are: (i) home directory, (ii) cropping size of animals in the videos, (iii) path to video files, etc. All the parameters are saved in config.yml.
Select the "configure" option from drop down menu and press "run" button. It will prompt you to enter the path of the MOTHe directory, select the directory and confirm. Next, it will ask you to select an example video. You can select any video from your video database, or one of the sample videos you have downloaded from here.
The next prompt is to enter the threshold values and step size for detection and tracking. The min and max threshold values depends on the specific case study (contrast between animal and background) and may take a few trial and error attempts to get it right (Read the section Choosing color threshold for more details).
You will also specify a step size (no. of frames to skip for detection and tracking task). If for any reason, you want to run the detection for every n frames instead of all the frames (it can speed up the detection task significantly). To track the video without skipping any frames, enter the step size as 1.
For learning/trial purposes, you may choose these values:
Wasp videos: min threshold = 150, max threshold = 250, step size = 50
Blackbuck videos: min threshold = 0, max threshold = 150, step size = 50
A window appears during the configuration process which is a frame of the test video you have chosen. This step is to determine the size of the animal to be detected and tracked. This value also helps in the dataset generation phase. Make sure to choose the most accomodating animal on the screen to avoid occlusions and missed detections later.
Click and drag across the animal to set the animal size for the configuration. Press the c key once to view the cropped animal. If satisfied with the click and drag process, proceed to press the c key again to confirm and end the configuration process. You can draw multiple boxes at a time and the last box will be taken to set the crop size. If you think that wrong box size got stored, you can repeat the configuration step from start.
You can view the config.yml file created in mothe folder.
You can now select the next option from drop-down menu of MOTHe-GUI to run data generation step.
Step 2: Data Generation
This program will pick frames from the videos, user can click on animals or background in these images to create samples for both categories (animal and background). Examples of animal of interest will be saved in the folder yes and background in the folder no.
User needs to generate at least 8k-10k samples for each category (see section How much training data do I need? for detailed guidelines). One must ensure to take a wide representation of forms in which animals appears in the videos and same for the background variation. However, for the trial purpose you can generate 5-10 images in each category.
MOTHe supports only binary classification. Therefore name the classes 'yes' for positive examples and 'no' for background examples. The data generation method takes a step size argument as well which helps the user to keep the number of examples per video in check. (Ex: a higher step size limits the number of frames per video. if a video is very long, one can set a higher step size to skip through unwated and consecutive frames). Please note that this step size is different from the one you entered in configure step, this step size allows you to generated data from widely spaced frames of the videos.
A window appears which is a frame of the video you have chosen. Start generating data at this point.
Click at the center of the animal once. The algorithm calculates the size of the bounding box based on the config file entry.
-Press the s key once to crop and store the animals once we have selected all the animals in the frame. Then it will take us to the next frame automatically.
-Press the n key to proceed to the next frame if the current frame is not worth collecting data from. Any selected animals are not cropped and stored if n key is pressed. It just takes us to the next frame.
-Press the u key if you want to undo a perticular selection that you have made.
-Once you are done collectiong samples from a video, press esc key to complete the process for this video.
Make sure to repeat this process for the 'no'.
Select all background examples in this case, i.e. background objects, habitat, any other category of objects present in the image.
At this point you will have two class folder with many examples to train the neural network.
You can close the data generation window now and proceed to the next step.
Step 3: Training the CNN
To train the neural network, select the "train" option from the drop down menu-
You can see the status of this step on the terminal/shell.
After successfully training the model, two graphs appear on the screen. The loss graph starts at a higher point and if the correct learning rate is applied, it takes a drastic decline and starts to plateau out as it reaches near zero. If a very high learning rate is applied, the graph starts travelling upwards instead of downwards. If a slightly higher learning rate is applied, it will not reack a closer point towards the zero line. The accuracy curve should travel upwards sharply and plateau out. It is important to avoid over fitting of data. This can be done by using adequate variance in the examples we generate during data generation. It is also important not to have too much variance since the accuracy may go down even though the network can generalize fairly well.
After training, the model gets stored in the mothe directory as mothe_model.h5py. This model will be used to detect and track animals in the test videos.
Step 4: Object detection
This step will detect the animals (object of interest) in the video frames. As mentioned earlier, this process is done in two steps - first the code predicts the areas in which animal may be potentially present (localisation) and then these areas are passes to the network for classification task. For localisation, we need thrsholding approach which gives us regions which have animals as well as background noise.
Initiate the detection process by selecting 'detection' option and pressing the run button. It will prompt to enter the name of a test video and the model which you want to use. You can use the already trained model available in MOTHe Github repository to run detection on blackbuck or wasp videos.
You can see the status in the terminal/shell.
After the successful detection, a video with detections and .csv are generated in the mothe folder.
STep 5: Object tracking
This step is used to ascribe unique IDs to the detected animals and it gives us the trajectories of the animals. It will use the detections from previous step. Hence, the input for this step would be original video clip and .csv generated in the previous step.
Initiate the tracking process by selecting "track" option. It will prompt to enter the name of a test video and the model which you want to use (preferably the same model you selected for the detection step).
You can see the status in the terminal/shell.
After the successful tracking, a tracked video and .csv are generated in the mothe folder.
This completes all the steps associated with animal detection and tracking in the videos. Once you have tested it on sample videos, you can use MOTHe for your particular dataset. Below are some guidelines which will help you select parameters for your videos.
MOTHe uses a CNN which uses a set of labelled examples to learn the features of the objects. Neural Networks generally work well with huge number of training samples. We recommend using at least 8-10k image examples for the animal category. This number may need to be increased if the animal of interest shows a lot of variation in morphology. For example, if males and females are of different colors, it is important to include sufficient examples for both of them. Similarly, if the background is too heterogeneous then you may need more training data (around 1500-2000 samples for different types of variations in the background). For example to train the MOTHe on our blackbuck videos, we used 9800 cropped samples for blackbuck (including males and females) and 19000 samples for the background because background included grass, soil, rocks, bushes, water etc.
The object detection steps requires user to enter threshold values in the config files. Object detection in MOTHe works in two steps, it first uses a color filter to identify the regions in the image on which to run the classification. We use color threshold to select these regions. You can see the values of thresholds for blackbuck and wasp videos in the config.yml file. If you are running MOTHe on a new dataset, there are two ways to select appropriate threshold values:
-
You may open some frames from different videos in an interactive viewer (for example MATLAB image display), you can then click on the pixels on animal and check the RGB values (take the avergae of all 3 channels). Looking at these values in multiple instances will give you an idea to choose a starting minimum and maximum threshold. Run the detection with these thresholds and you can improve the detection by hit and trial method to tweak the threshold.
-
You can compare your videos to wasp and blackbuck videos and start with threshold values to which your data is more similar. For example, if your animal looks more similar to blackbuck in color and lighting conditions, you may start with default thresholds and improve the detection by changing lower and upper threshold by little amount at a time.
-
We have also added a functionality with which you can play around with threshold values to see how different threshold values converts the image. Consider trying to determine the threshold value to suit your data. At the start of application, press the 'Determine threshold' button and play with the slider bar (0-255) to determine the approximate values between which the individual may be considered to be a keypoint by the blob detector.
##Referencce:
If you find this package interesting, relevant or use for your work (with or without modifications), please cite the following manuscript:
Rathore A, Sharma A, Shah S, Sharma N, Torney C, Guttal V. 2023. Multi-Object Tracking in Heterogeneous environments (MOTHe) for animal video recordings.” PeerJ 11:e15573 https://doi.org/10.7717/peerj.15573 (also at bioRxiv: https://www.biorxiv.org/content/10.1101/2020.01.10.899989v2.full)
