The BAMBI Matched Dataset: Raw Nadir UAV RGB/Thermal Recordings of Forest Wildlife with Red Deer MOT Labels (Part 2)

This dataset is part of the BAMBI (Biodiversity Airborne Monitoring Based on Intelligent UAV sampling; https://www.bambi.eco/) project. It contains the raw, unrectified nadir UAV recordings of forest wildlife as extracted video frames in two spatially matched modalities, together with reviewed multi-object tracking (MOT) labels for red deer. Each flight is stored as an individual ZIP archive named by its numeric flight ID (e.g. flight_1.zip). Inside each ZIP: rgb/ – extracted RGB frames (<frame>.jpg) thermal/ – extracted thermal frames (<frame>.jpg) labels/<flight>_rgb_mot.txt, labels/<flight>_thermal_mot.txt – MOT boxes in source-video pixel coordinates, one row per frame,track_id,x,y,w,h,conf,-1,-1,-1, already trimmed to the review labels/<flight>_labels.json – per-track gender (male | female_juvenile | unknown | null), occlusion (per-modality frame ranges), class_name, accepted_modalities, trim_start_frame/trim_end_frame and label_source manifest.json – frame counts and label provenance for the flight Frame file names are the source-video frame index, zero-padded, and are shared across modalities: rgb/000010.jpg and thermal/000010.jpg are temporally aligned. Frames were extracted at a fixed stride, so the frame index advances in steps rather than one-by-one. Note on MOT frame indices: the frame column of the MOT files is a source-video frame index at full video rate and therefore names many frames that were not extracted as images. To align a MOT row with an image, match its frame value against the extracted frame file names; rows whose frame was not extracted have no corresponding image. Per-frame occlusion is obtained by testing whether a MOT row's frame falls inside one of its track's occlusion ranges for that modality. Tracks with gender = null are accepted non-red-deer or unlabeled tracks. Dataset-wide all_track_labels.json (every flight's track labels in one file), summary.json (per-flight counts) and README.md are provided as standalone downloads in each part. MOT labels cover the curated red deer flights; flights outside that scope are included as frames only, without a labels/ folder. Due to the size of the full dataset it has been split into multiple parts. Please download all parts to obtain the complete dataset. This is a subset of 'The Bambi Dataset: Multimodal Nadir UAV-Recordings of Forest Wildlife' which can be accessed via https://github.com/bambi-eco/Dataset.