HotSight / Data /README.md
hotsight's picture
Fix relative paths after the PseudoLabeling/ move and stale split-file names in docs
490858f verified
|
Raw History Blame Contribute Delete
2.31 kB

Sampled dataset

Data/ contains 951 synchronized RGB/IR frame pairs sampled from Canopy and Graminoid flights, normally every 300 RGB frames (approximately every 10 seconds). Each clip has this layout:

<unit>/<clip>/
├── rgb/fXXXXX.png
├── ir/fXXXXX.png
├── ir/fXXXXX_recovered.png   # only where recovery was needed
├── mask/fXXXXX.txt
├── mask/fXXXXX_overlay.jpg   # inspection only
└── homography/{homography.json,points.json}

fXXXXX is the original zero-based RGB frame index. RGB images are 2688×1512 and IR images are 640×360. Each mask is a flattened 2688×1512 row-major array of class labels 1–5. samples.csv lists the sampled pairs. splits_graminoid.csv (Graminoid held out as the test set) and splits_canopy.csv (Canopy held out) give each frame's train, val, or test assignment in the two evaluation settings, and annotation_label_by_split_graminoid.csv / annotation_label_by_split_canopy.csv the class pixel shares per split.

Known gaps

  • Canopy-19 is a 0.2-second dead clip, and Graminoid-03 has only six seconds of IR; both are empty because sampling starts near the 10th second.
  • Graminoid-11 and Graminoid-14 have no points.json; their homographies were copied from the immediately preceding clip in the same flight, as noted in each homography folder.
  • There are 87 _recovered.png IR alternatives (for IR/RGB blended clips). Use these in place of the same-stem original when present; they are not referenced separately in samples.csv.

Creating dataset splits

Use scripts/make_random_splits.py to hold out one burn site as the test set and divide the other site into training and validation sets.

The script writes Data/splits.csv with one row per sampled frame and the columns frame and split. Split values are train, val, or test. The released splits_graminoid.csv and splits_canopy.csv are that output for --test Graminoid and --test Canopy, renamed.

The related clips documented in markdown/flights.md are treated as one inseparable flight block, so they cannot be divided between training and validation. The script randomly selects, using seed 42 by default, from all complete-block combinations that place 10–20% of the non-test frames in validation.