gauravkaul/DAMM_mouse_detection

Datasets used to create the Detect Any Mouse Model (or DAMM for short )

DAMM is a project focused on the detection and tracking of multiple animals within complex social and environmental settings. The goal of this project was to create mouse instance segmentation systems that generalize across cage and experimental setups. This README provides an overview of the datasets used in DAMM and the annotation structure.

Annotation Structure

Annotations in this project are structured using a JSON format compatible with COCO-style annotations. Each annotation corresponds to an object in an image and includes the following fields:

• file_name: The name of the image file, including its path.
• height: The height of the image in pixels.
• width: The width of the image in pixels.
• image_id: An identifier for the image.
• annotations: A list of annotations, where each annotation (dictionary) contains the following fields:
  • bbox: A bounding box representing the object's location in the image. Expressed as [[x1, y1], [x2, y2]], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.
  • bbox_mode: The format of the bounding box coordinates, specified as BoxMode.XYXY_ABS for absolute coordinates in XYXY format.
  • category_id: The category identifier for the object.
  • segmentation: A list of closed polygon regions representing the object's mask. Each region is represented as [x1, y1, x2, y2, ..., xn, yn].

Detection Datasets

Detection datasets used in DAMM are stored in the detection_datasets/ directory (unzip detection_datasets.zip). The following tables list the number of examples for each split and their associated folder location. Each folder contains a metadata.json file containing the annotations.

AER Pretraining Dataset

Dataset Name	Number of Examples	Folder Location
AER pretraining	2209	AER_pretraining_set/

AER Lab Sourced (In-house) Datasets

Dataset Name	Number of Examples	Folder Location
playground	103	playground/
light_dark_nest	104	light_dark_nest/
perfect_setup	102	perfect_setup/
colorful_multi_mice	102	colorful_multi_mice/
black_multi_mice	101	black_multi_mice/

Web Sourced (Publicly Available) Datasets

Dataset Name	Number of Examples	Folder Location
golden_open_field	101	golden_open_field/
mars_multi	104	mars_multi/
mcdannald_rat_operant_chamber	106	mcdannald_rat_operant_chamber/
trimice_dlc	101	trimice_dlc/
pennington_maze	102	pennington_maze/
golden_home_cage_color	100	golden_home_cage_color/
golden_home_cage_greyscale	101	golden_home_cage_greyscale/

Challenge Datasets (Located within the challenge_setups/ folder)

Dataset Name	Number of Examples	Folder Location
top_black_DSLR	71	top_black_DSLR/
ratcage_brown_DSLR	73	ratcage_brown_DSLR/
ratcage_white_actioncam	74	ratcage_white_actioncam/
ratcage_black_actioncam	74	ratcage_black_actioncam/
grate_brown_actioncam	73	grate_brown_actioncam/
grate_white_actioncam	71	grate_white_actioncam/
top_brown_DSLR	71	top_brown_DSLR/
ratcage_black_DSLR	75	ratcage_black_DSLR/
grate_white_DSLR	71	grate_white_DSLR/
ratcage_brown_actioncam	75	ratcage_brown_actioncam/
grate_black_actioncam	71	grate_black_actioncam/
grate_brown_DSLR	75	grate_brown_DSLR/
top_white_DSLR	100	top_white_DSLR/
top_brown_actioncam	72	top_brown_actioncam/
grate_black_DSLR	71	grate_black_DSLR/
top_white_actioncam	70	top_white_actioncam/
top_black_actioncam	71	top_black_actioncam/
ratcage_white_DSLR	71	ratcage_white_DSLR/

---

Tracking Datasets

Tracking datasets used in DAMM are stored in the tracking_datasets/ directory (unzip tracking_datasets.zip). Each folder contains tracking data similarly structured to detection data, with the additional key of frame_id for each annotation mapping to the video frame number.

Single Animal Datasets (Located within the single_animal_tracking/ folder)

Dataset Name	Frames Annotated	FPS	Duration (seconds)
oft_12/oft_12.mp4	1084	25.0	43.36
mcdannald_rat_operant_chamber_conditioning_02/mcdannald_rat_operant_chamber_conditioning_02.mp4	828	14.80694563265613	56.730133333333335
golden_mouse_open_field_social_familiarity_cd1_absent_01/golden_mouse_open_field_social_familiarity_cd1_absent_01.mp4	1387	30.0	46.233333333333334
epm_1/epm_1.mp4	2410	24.99810498811716	157.691953125
single_spinach/single_spinach.mp4	1531	30.0	51.03333333333333
smear_mouse_operant_chamber_olfactory_search_raw_01/smear_mouse_operant_chamber_olfactory_search_raw_01.mp4	145	10.0	14.5
pennington_mouse_operant_chamber_exploration_02/pennington_mouse_operant_chamber_exploration_02.mp4	1514	25.0	60.56

Multi-Animal Datasets (Located within the multi_animal_tracking/ folder)

Dataset Name	Frames Annotated	FPS	Duration (seconds)
multi_ir_lighting/multi_ir_lighting.mp4	365	9.912283083942102	36.823
golden_mouse_operant_chamber_social_self_admin_09/golden_mouse_operant_chamber_social_self_admin_09.mp4	437	30.0	52.166666666666664
mouse024_task1_annotator1/mouse024_task1_annotator1.mp4	1965	30.0	65.5
golden_mouse_home_cage_social_72/golden_mouse_home_cage_social_72.mp4	1470	30.0	49.0
top_white_iphone_3/top_white_iphone_3.mp4	174	29.97	75.00834167500834

---

This format maintains clarity while indicating the folder locations for each dataset category.

Usage and Citation

If you use this dataset, please make sure to cite the following article:

@article{kaul2024damm,
  author    = {Gaurav Kaul and Jonathan McDevitt and Justin Johnson and Ada Eban-Rothschild},
  title     = {DAMM for the detection and tracking of multiple animals within complex social and environmental settings},
  journal   = {bioRxiv},
  year      = {2024}
}

license: cc-by-4.0