Experimental Result

Track 1: Uniform Split
- Range View
- Voxel
Track 2: Sequential Split
- Range View
- Voxel

Track 1: Uniform Split

This track is analogous to the semi-supervised image segmentation community, which sample LiDAR scans with a uniform probability.

Range View

Method	nuScenes					SemanticKITTI					ScribbleKITTI
Method	1%	10%	20%	50%	Full	1%	10%	20%	50%	Full	1%	10%	20%	50%	Full
RangeNet++
w/ LaserMix
improv. ↑	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺
Download		[link]			[link]		[link]			[link]		[link]			[link]

SalsaNext
w/ LaserMix
improv. ↑	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺
Download		[link]			[link]		[link]			[link]		[link]			[link]

FIDNet	38.3	57.5	62.7	67.6	72.1	36.2	52.2	55.9	57.2	61.4	33.1	47.7	49.9	52.5
w/ LaserMix	49.5	68.2	70.6	73.0		43.4	58.8	59.4	61.4		38.3	54.4	55.6	58.7
improv. ↑	⁺11.2	⁺10.7	⁺7.9	⁺5.4	⁺	⁺7.2	⁺6.6	⁺3.5	⁺4.2	⁺	⁺5.2	⁺6.7	⁺5.7	⁺6.2	⁺
Download		[link]			[link]		[link]			[link]		[link]			[link]

CENet	40.4	58.7			71.7	40.4	56.0	58.2	60.1	62.9
w/ LaserMix
improv. ↑	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺
Download		[link]			[link]		[link]			[link]		[link]			[link]

Voxel

Method	nuScenes					SemanticKITTI					ScribbleKITTI
Method	1%	10%	20%	50%	Full	1%	10%	20%	50%	Full	1%	10%	20%	50%	Full
MinkUNet		69.4			72.8					62.8
w/ LaserMix
improv. ↑	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺
Download		[link]			[link]		[link]			[link]		[link]			[link]

SPVCNN	53.7	69.7	71.6	72.0	73.2	42.7	57.5	60.2	62.7	63.2	38.2	53.8			56.7
w/ LaserMix										65.8
improv. ↑	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺2.6	⁺	⁺	⁺	⁺	⁺
Download		[link]			[link]		[link]			[link]		[link]			[link]

Cylinder3D
w/ LaserMix
improv. ↑	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺	⁺
Download		[link]			[link]		[link]			[link]		[link]			[link]

Track 2: Sequential Split

This track takes into account the LiDAR data collection nature when sampling LiDAR scans.

Range View

Available soon.

Voxel

Available soon.