Generating features using APTProject issue

[dnaligase]

Hi everyone!

I wondered if I could convert data from DeepLabCut (DLC) that I've been using to analyze side-view single mouse video to JAABA to classify the behavior. I found that it is possible to generate features using the built-in "Import from APT" option, which utilizes an APT .lbl file. So I decided to make a simple script that would convert DLC .csv to .trk (it creates and replaces pTrk and pTrkFrm variables of already existent .trk obtained from APT).

So what I do is import labels from my newly created .trk via APT, then save the project as .lbl. After that, I try to create features using the aforementioned (though the figure doesn't display correctly, as in the screenshot) and create the project. However, I'm not able to create the experiment and I run into a warning message as shown below. I see it's a pretty messy way to do it, but what is it I'm doing completely wrong here and is there any way to work with JAABA having only trajectories?

Thank you for your help,
Georgy

[mkabra]

Hi Georgy, That's not an incorrect way to import DLC projects into APT, but there is already a script in APT repository to import your existing projects ( kristinbranson/APT#309 (comment)) which will make it easier for you to import the project. As for the messed up layout, I've pushed a fix. Let me know if it still doesn't look right the next time you open it. This happens often in Matlab when developing GUIs on Linux and using it on Windows. The warning message window shouldn't have been that big. Can you tell me what version of Matlab are you using? The warning message is expected, but do you not see any buttons at the end of the window? You should click on "Generate.." button or something similar to that if you see the buttons. Mayank

…

On Sun, Apr 19, 2020 at 4:06 AM dnaligase ***@***.***> wrote: Hi everyone! I wondered if I could convert data from DeepLabCut (DLC) that I've been using to analyze side-view single mouse video to JAABA to classify the behavior. I found that it is possible to generate features using the built-in "Import from APT" option, which utilizes an APT .*lbl* file. So I decided to make a simple script that would convert DLC .*csv* to .*trk* (it creates and replaces pTrk and pTrkFrm variables of already existent . *trk* obtained from APT). So what I do is import labels from my newly created .*trk* via APT, then save the project as .*lbl*. After that, I try to create features using the aforementioned (though the figure doesn't display correctly, as in the screenshot) and create the project. However, I'm not able to create the experiment and I run into a warning message as shown below. I see it's a pretty messy way to do it, but what is it I'm doing completely wrong here and is there any way to work with JAABA having only trajectories? Thank you for your help, Georgy [image: APTProject] <https://urldefense.com/v3/__https://user-images.githubusercontent.com/60392997/79672065-6921dc80-81d7-11ea-8ec5-1224e7d6aa07.PNG__;!!Eh6p8Q!Q8aUYAKE8q7CTqTGtd4CcCLsClX0ZdD1966driIWDc10E0zfi24w0TiLuiXAg6l3Db8$> [image: Warning] <https://urldefense.com/v3/__https://user-images.githubusercontent.com/60392997/79672310-60caa100-81d9-11ea-9e72-ae041d0c85d0.PNG__;!!Eh6p8Q!Q8aUYAKE8q7CTqTGtd4CcCLsClX0ZdD1966driIWDc10E0zfi24w0TiLuiXAzDbHnKc$> — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <https://urldefense.com/v3/__https://github.com/kristinbranson/JAABA/issues/332__;!!Eh6p8Q!Q8aUYAKE8q7CTqTGtd4CcCLsClX0ZdD1966driIWDc10E0zfi24w0TiLuiXAY3-cdLc$>, or unsubscribe <https://urldefense.com/v3/__https://github.com/notifications/unsubscribe-auth/AAJNKY2CB67RDR5PZDY2IUTRNITPHANCNFSM4MLRAZWA__;!!Eh6p8Q!Q8aUYAKE8q7CTqTGtd4CcCLsClX0ZdD1966driIWDc10E0zfi24w0TiLuiXA9IszCuM$> .

[dnaligase]

Dear Mayank,

Thank you for your reply! Layout now looks great, and the function you proposed also works (though it uses white space delimiter by default which may cause some troubles with train data when it contains ' ' in video names, but it's quickly fixable). I guess now I should retrain the network. I'll come back as soon as it's done.

Regarding the warning window: it does display the 'OK' button, but nothing happens when clicked. My Matlab version is R2018b.

According to your reply, I assume that one .trk alone converted from .csv (containing all DLC tracks for each body part) allows generating features skipping the step of retraining in APT...? That is, in .lbl file, the way of getting which I described above, I have only labels, extracted from manually created .trk, whereas other information is missing (e.g. all trained trackers, locations of videos used for training, etc.). Is this important for using APTProject?

Kind regards,
Georgy

[mkabra]

Hmm, I think the label file that you created could work. Not entirely sure though. For one, APTProject reads a labeled frame for display. But you've probably gotten around it because I don't see it in the snapshot. I don't remember other dependencies in the code though. I guess you could try it out and see if it works. However, the code does depend a lot on the structure of the .trk file. It needs to have the .trk_info field which has information about crop locations (APT allows you to do tracking in a portion of the video instead of the whole video), and video frame size for computing the features and other stuff. And I think it'll be difficult to get a lot of those right if you do the conversion manually. If you'd want to continue using your existing DLC tracking output, what you can do is create a lbl file, train a dummy tracker (say train it only for 100 iterations), track using this dummy tracker, and then replace the *pTrk *in *.trk* with DLC output. Which honestly is quite convoluted and so you might as well train the tracker in APT. So it doesn't do anything when you hit "OK" in the warning window? Can you take a look at Matlab's command window? It might have printed errors and such there. Mayank

…

On Wed, Apr 22, 2020 at 3:28 AM dnaligase ***@***.***> wrote: Dear Mayank, Thank you for your reply! Layout now looks great, and the function you proposed also works (though it uses white space delimiter by default which may cause some troubles with train data when it contains ' ' in video names, but it's quickly fixable). I guess now I should retrain the network. I'll come back as soon as it's done. Regarding the warning window: it does display the 'OK' button, but nothing happens when clicked. My Matlab version is R2018b. According to your reply, I assume that one .*trk* alone converted from . *csv* (containing all DLC tracks for each body part) allows generating features skipping the step of retraining in APT...? That is, in .*lbl* file, the way of getting which I described above, I have only labels, extracted from manually created .*trk*, whereas other information is missing (e.g. all trained trackers, locations of videos used for training, etc.). Is this important for using APTProject? Kind regards Georgy — You are receiving this because you commented. Reply to this email directly, view it on GitHub <https://urldefense.com/v3/__https://github.com/kristinbranson/JAABA/issues/332*issuecomment-617435503__;Iw!!Eh6p8Q!XamHg1FnX_00NdI2PJSw1BJPybfaK0T7ykrUPl9MFgcGFJ8xSTtN4Qs72IeaR52hurE$>, or unsubscribe <https://urldefense.com/v3/__https://github.com/notifications/unsubscribe-auth/AAJNKY7ZYS2AZQ64IV7ZLETRNYJIDANCNFSM4MLRAZWA__;!!Eh6p8Q!XamHg1FnX_00NdI2PJSw1BJPybfaK0T7ykrUPl9MFgcGFJ8xSTtN4Qs72IeaPvvHm5w$> .

[dnaligase]

Dear Mayank,

Here's the command window output attached. I suppose it'd be much simpler to just use APT in a way you provided.

Thanks a lot,
Georgy