about the labels of suncg data

[KnightOfTheMoonlight]

Dear Shuran,

I tried to use your great dataset as the input of my own network. But there's one issue I encountered recently, I find there's one sample that has labels [ 0, 1, 2, 3, 4, 7, 13, 14, 18, 27, 37] This is so weird, from your matlab_code, I think the labels for suncg dataset is [0,1, ... 36], and you map them into 11 classes.

[KnightOfTheMoonlight]

Here are the training samples that I found at present, the label is out of range.

SUNCGtrain_3001_5000/00016521_17da926b54ab4a91126d43718dc2e9a2_fl001_rm0006_0000.bin
SUNCGtrain_3001_5000/00016524_17da926b54ab4a91126d43718dc2e9a2_fl001_rm0006_0000.bin

[KnightOfTheMoonlight]

And I found that in your code, actually you just ignore this by c++ language rules.
From this function ReadVoxLabel

for (size_t i = 0; i < scene_vox_RLE.size() / 2; ++i) {
unsigned int vox_val = scene_vox_RLE[i * 2];
unsigned int vox_iter = scene_vox_RLE[i * 2 + 1];
for (size_t j = 0; j < vox_iter; ++j) {
if (vox_val == 255) {
segmentation_label_fullsize[vox_idx] = Dtype(255);
occupancy_label_fullsize[vox_idx] = Dtype(0.0f);
} else {
segmentation_label_fullsize[vox_idx] = Dtype(
segmentation_class_map[vox_val]);
if (vox_val > 0)
occupancy_label_fullsize[vox_idx] = Dtype(1.0f);
else
occupancy_label_fullsize[vox_idx] = Dtype(0.0f);
}
vox_idx++;
}
}

Acturally even if vox_val is over 36, segmentation_class_map[vox_val] can still return 0, instead of error report.