Dataset Viewer
Auto-converted to Parquet Duplicate
Search is not available for this dataset
image
imagewidth (px)
360
5.57k
label
class label
3 classes
0explainer
0explainer
0explainer
0explainer
0explainer
0explainer
0explainer
0explainer
0explainer
0explainer
0explainer
0explainer
0explainer
0explainer
0explainer
0explainer
0explainer
0explainer
0explainer
0explainer
0explainer
0explainer
0explainer
0explainer
0explainer
1final
1final
1final
1final
1final
1final
1final
1final
1final
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
2raw
End of preview. Expand in Data Studio

Nikon Z 30 handheld lunar burst, October 26, 2026

A handheld burst of the Moon shot by Khang Ly with a Nikon Z 30 and the Nikkor Z DX 50–250 mm at 250 mm, ISO 100, 1/250 s, 14-bit lossless NEF with a camera JPEG beside each frame, Continuous H (extended) release. The purpose was to test whether a cheap APS-C body and a kit telephoto can be stacked into a sharper, cleaner Moon than any single frame, with no tripod and no tracking.

Contents

  • raw/: 173 NEF + JPEG pairs (DSC_7956 to DSC_8128), about 20 MB per NEF, straight from the card. The frames move by about 5 px between shots and must be aligned before any sharpness judgement.
  • final/: the stack outputs. moon_stack_70pct_1.5x_linear.tif is the linear-light Bayer-drizzled stack of the sharpest 70 % of frames at 1.5× scale; the moon_final_* images are three renders of it (neutral, daylight white balance, and a saturation-boosted "mineral" version); compare_crop.jpg and tycho_single_vs_stack_2x.jpg compare the best single frame with the stack.
  • explainer/: the figures of the accompanying guide, including stack_NNN.jpg, the stack after N frames from 1 to 108, and data.json with the per-stage numbers.

What was found

155 frames went into the stack and 108 passed the sharpness grade. Noise fell as √N (7.1× at 54 frames against 7.3× in theory), and the clean stack took sharpening a single frame cannot: limb blur 0.64 px against 0.83 px for the best single frame. Two pipeline notes that cost time: OpenCV's phaseCorrelate with a window writes into its inputs when they are views of a larger array, and rawpy's color_matrix is zeros until postprocess() has run, so the camera-to-sRGB matrix must be built from rgb_xyz_matrix as dcraw does.

Also worth knowing for anyone stacking Z 30 frames: the camera's NEF is spatially filtered from 1/4 s upwards even with long-exposure noise reduction off (single-pixel stars and hot pixels are clipped to their neighbourhood), while frames of 1/6 s and shorter, like these, are untouched. See the companion dataset khangly/z30-dark-frames.

Use

Load the NEFs with rawpy (raw_image_visible, black level 1008 on every channel, white level 16,383). The stacking pipeline (stack.py: grading, phase-correlation alignment on copies, Bayer drizzle, linear merge) is in the camera repository the dataset belongs to.

Licence: CC BY 4.0. Photographs by Khang Ly.

Downloads last month
268