Knowledge Base

Adjust the Settings for Nose-Tail Base Detection (Deep Learning)

Aim

To achieve detection of the nose- and tail-base points of your subjects using the Deep Learning method. This topic applies to tracking of one subject per arena. If you track two or more subjects per arena with Deep Learning, you do not need to adjust settings besides the sample rate.

Prerequisites

• In the Experiment Settings:
  • Under Subjects, 1 is selected.
  • Under Tracked Features, Center-point, nose-point and tail-base detection is selected.
  • Under Body Point Detection Technique, Deep Learning is selected.
• You optimized the lighting and the background based on Deep Learning: Requirements.

Procedure

1. Open the Detection Settings.
2. In the Video section (top-right), select the video file (if you track offline) and choose the sample rate. See Video File, Image Quality and Sample Rate.
3. Under Advanced > Method, choose the detection method and the contrast range. See Advanced Detection Settings: Method.
4. In the Detection Settings window, under Detection, click the Automated Setup button and follow the instructions. When the Automated Setup gives good detection, proceed with the next step. See Detection Settings: Automated Setup.
5. Play the video or wait until the subject in the live image is located (a) far from walls and other objects (especially when the walls and objects provide little or no contrast with the subject), (b) its body is not curled or contracted, and (c) its nose is visible. For best results, position the video on a frame where the animal is slightly stretched.
6. If you work with hooded animals, like the Lister rats or the Long-Evans rats, under Method, next to Deep Learning Settings, select Hooded Rats.
7. Under Method, next to Deep Learning Settings, click the Define button. The Cutout window opens. EthoVision XT shows a square box around the subject. The size of this box should already be optimized based on the detected subject. The Cutout box should include the whole subject's body leaving some space around it.
8. If that is not the case, click Automated.
9. Click OK.
10. The Video window now shows the subject with its nose and tail-base highlighted (nose = light blue).
11. If the nose and tail-base are not detected sufficiently well, click Define again and move the slider until the box includes the entire body of the subject. See the suggestions below.

• The tail of the rodent does not have to be included.
• Important: Keep some space between the subject and the outline of the Cutout box.

Notes

• To improve detection, see also Deep Learning: Requirements.
• The Cutout box size is saved in the Detection Settings. The next time you click Define, the Cutout dialog shows the last saved value.
• Which Value for the Cutout Box Size? The Cutout value is shown for reference. Do not focus on a specific value, because other similar values may work fine (for example, 135 and 137). However, take note of that value if you know that it works, so you can use it in the next experiment assuming that you use the same camera distance, arena size, etc. As a rule of thumb, the Cutout box should include the animal also when it is stretched. Make sure that the box includes some space around the animal, at least half the body length.
• Small Animals: With small animals, the Cutout box can quickly become too small or too big. Adjust the Cutout value by small steps, until the nose and tail-base points are detected correctly.
• Occlusions: When the subject image is obstructed by, for example, a door, increase the Cutout box size so that the nose or tail can be found on the other side of the obstruction. This usually improves the detection.
• Reflections: If there are reflections at the wall of the apparatus and these give problems with detection of the nose point, try to reduce the size of the Cutout box. See an example in Troubleshooting: The Detected Nose Point Is Far from the Animal's Contour.
• Multiple Arenas: When you work with two or more arenas, the Cutout box is displayed only in the last arena. Adjust its size as described above.

Examples of a Good Cutout Box Size

Subject in an Open Field: In the first picture below, detection won't work because the rat's nose is outside the box. In the second picture, detection may work but a large Cutout size increases the risk that other objects are included that are of similar color as the rat.

Detection Settings for Behavior Recognition

Prerequisites

• You have the Rat/Mouse Behavior Recognition Module.
• In the Experiment Settings, under Tracked Features, you have selected Center-point, Nose-point and tail-base tracking. Under Analysis Options, you have selected Behavior recognition.
• See also Behavior Recognition: Requirements

Procedure

1. Put the animal in the arena or play the video. Optimize the camera setup, and lighting conditions. Try to reduce reflections on the walls of the arena as much as possible.
2. Select the image source and sample rate. In the Detection Settings pane, under Video, select a value of sample rate between 25 and 30 samples/second. See Video file, image quality and sample rate.
3. Use the Automated Setup as described in Detection settings: Automated setup.
4. When detection is good, open the Behavior Recognition section in the Detection Settings pane and follow Advanced detection settings: Behavior recognition.

Notes

• If detection with the Automated Setup is not good, in the Detection Settings pane, click Detection and then Advanced. Under Method, select one of the following:
  • Rodents / Default. This is selected automatically when you use Behavior recognition, and should work in most cases.
  • Rodents / For occlusions. Choose this method only when Default does not give good results. This option is optimal when there are objects in the arena (for example a novel object) which create occlusions and make detection of the whole subject's body difficult.
• The quality and amount of light is very important when using Behavior recognition. See Subject exposure.