idtrackerai

FAQs

General questions

Can I use idtracker.ai in my videos?

You can check our Gallery to see the type of videos in which idtracker.ai worked well. We also give a set of General video conditions that we advise users to follow to get the best results with idtracker.ai.

Does idtracker.ai work in Windows?

Yes, in the Gitlab Repository we provide instructions to install idtracker.ai in Windows. We have tested the installation in computers running Windows 10 Pro.

Can I run idtracker.ai in a laptop?

Yes. We are running idtracker.ai with all its features in high-end performance gaming laptops from Obsidian. If your laptop does not have a GPU you can still use idtracker.ai, see the previous FAQ.

Can I use idtracker.ai if my computer does not have a good GPU?

Yes, you can still use idtracker.ai if you don’t have a GPU. However, the parts of the tracking that use the GPU will be up to 100x slower. However, if you are tracking a single animal, or if you are tracking groups of animals but you do not care about their identities, you can use idtracker.ai and track animals at a faster speed. To track single animals read the previous FAQ. To track animals without keeping their identity, just click the button Tracking without identities when you get to the Tracking tab of idtracker.ai (after the video is segmented).

Can I use idtracker.ai in an AMD GPU?

Althought there are some efforts on making Tensorflow work with AMD GPUs, we do not provide support or installation instructions in this case. As idtracker.ai is free and open source, you can always download it and set your environment with Tensorflow supporting your AMD GPU. However, we have not heard from any user using idtracker.ai with AMD GPUs, so if you are successful, please let us know!

Can I use idtracker.ai with Google Colab?

Currently idtracker.ai is a GUI (Graphical User Interface) based software. This means that you cannot easily run it from a terminal. As idtracker.ai is free and open source, you can always modify the code to run the GPU intensive parts from a terminal and hence from Google Colab.

We are currently working on a new GUI and code structure using Pyforms that will be released by mid-2019. This will allow users to interact with idtracker.ai using the terminal. Then we will provide a Jupyter Notebook with the necessary commands to run idtracker.ai in Google Colab.

Is there a Docker image for idtracker.ai?

Currently idtracker.ai is a GUI (Graphical User Interface) based software. Docker typically works easily when you can run the software from the terminal (command line). Thus, for now it does not make sense to provide a Docker image.

However, we are working on a new GUI and code structure using Pyforms that will be released by mid-2019. This will allow users to interact with idtracker.ai using the terminal. Then we will provide a Docker image that you will be able to use to run idtracker.ai from the terminal.

Can idtracker.ai track multiple videos in batch?

Currently to track every video you need to do it from the idtracker.ai GUI. In the near future we will release a version of idtracker.ai where you will be able to set the preprocessing parameters for multiple videos and then use a bash script to tell idtracker.ai to track those video, either in your computer or in a cluster.

Does idtracker.ai give orientation and posture information?

Orientation and posture can be computed a posteriori from the blobs_collection.npy file that idtracker.ai generates. In this repository we provide an example where we compute the nose, tail and midline for fish.

You can also generate a small video for every animal in the group and use it to get the posture with one of the AI based posture trackings (Deeplabcut, LEAP, …).

Can I use idtracker.ai together with DeepLabCut or LEAP?

As explained before, you can generate a small video for every animal in the group and use it to track the body parts with one of the AI based posture trackings (Deeplabcut, LEAP, …).

Does idtracker.ai track single animals?

Yes. Although idtracker.ai is designed to track multiple animals keeping their identities along the video, you can also track videos with a single animal. When the system asks for the number of animals, just type \(1\) and follow the instructions. The system will automatically skip the GPU intensive parts that are not necessary in this case. This means that you can use idtracker.ai to track single animals in a desktop or a laptop computer even if it does not have a GPU. In this case, idtracker.ai will perform very fast as it will only need to segment the video to extract the position of yoru animal.

Can I track humans with idtracker.ai?

We haven’t tried to track people with idtracker.ai. We think that idtracker.ai can track well people in videos recorded under laboratory conditions. Tracking humans on natural environments (streets, parks,…) it is a much more difficult task for which idtracker.ai was not designed. However, as idtracker.ai is free and open source, you can maybe use parts of our algorithm to set your human tracking for natural environments.

Common installation problems

Some of the errors that you might encounter might have been already reported by other users and fixed. Please update your idtracker.ai to make sure you are using the latest version. To update idtracker.ai follow the instructions in the idtracker.ai gitlab repository

If the error persists, please report the issue in the idtracker.ai gitlab repository or send us an email to idtrackerai@gmail.com. We will try to fix it as soon as possible.

Solving environment: failed

Solving environment: failed

ResolvePackageNotFound:
- libtiff==4.0.9=vc14_0

This error occurs when one of the libraries listed in the environment.yml file (in this case the libtiff library) has been updated in the Conda cloud repository and the version does not match. You can try to solve the error by checking which is the latest version in the Conda cloud.

Common GUI (Graphical User Interface) bugs and questions

We are constantly improving the GUI. However, you might still find some bugs, please report them. The following bugs that we describe do not affect the tracking performance, and you can still use idtracker.ai is you learn how to avoid them.

Reopening and validating a previous tracking session

idtracker.ai allows you to revisit previous tracking sessions to, for example, validate the trajectories of a tracked video, or to recompute some step of the tracking pipeline.

You can do this by starting idtracker.ai and selecting again the .avi file of the video that you want to revisit. When the GUI asks for the session name, input the name of the session you want to revisit (exclude the session_ part). A pop-up with the different steps computed will appear. For example, if you restore all the steps, a pop-up will appear asking if you want to go to the Global or Individual Validation tabs. By selecting one of them, a last pop-up will ask if you want to check the trajectories with the animals identified during the crossings or without identities during crossings. Press one of them to validate the trajectories.

Note that you can only modify the identities in the option “With animals not identified during crossings”. When you save the new modified identities the software will automatically reinterpolated the trajectories to identify the animals during the crossings.

Empty tabs

idtracker.ai has a very lineal processing procedure. If some tabs appear empty it is because you haven’t perform the step necessary to move to the next tab. For example, at the beginning only the Welcome tab will have content, the rest of the tabs will be empty. First you need to select a video for the ROI Selection and Preprocessing tabs to become active. The same will occur if you are in the preprocessing tab and try to go to the Tracking or Global Validation tabs. First you need segment the video for the tab Tracking to become active.

ROI warning popup but ROI is selected

When moving form the Preprocessing tab to the ROI Selection tab, and trying to save a ROI, a ROI warning might pop with the following message:

It seems that the ROI you are trying to apply corresponds to the entire frame. Please, go to the ROI Selection tab to select and save a ROI.

You should ignore it and press the save ROIs button again. Then your ROI will be saved.

idtracker.ai crashes when selecting video

If you select a video, the Session Name popup will appear for you to input the session name. If you go out of this popup and try to select the video again, you might get the following error.

File "/home/polaviejalab/idtrackerai/idtrackerai/gui/select_file.py", line 121, in open CHOSEN_VIDEO.set_chosen_item(self.filechooser.selection[0])
IndexError: list index out of range

Once you have selected a video, you should input the session name and pres ENTER. If you selected the wrong video, you should close the idtracker.ai GUI and open it again.