search

Weafre / MNeT

A Torch implementation of the scalable point cloud attribute compression method MNeT (ICASSP 2023)
2 stars 4 forks source link

No results consistent with the article data were obtained #2

Open kisworking opened 7 months ago

kisworking commented 7 months ago

Hello, may I ask what are the specific steps for processing datasets? Do we need to process all the ply files for each dataset? I have trained on datasets other than those in the paper table and CAT1, but after following the instructions in readme, I cannot obtain the corresponding results. At the same time, are the specific settings for other parameters set by default? What should be done when preprocessing CAT1 and owlii with a large number of files?

zb12138 commented 6 months ago

Same as mine. I got loot_vox10_1200 49bpp

Encoded file:  datasets/MPEG/MPEGCat1A/loot_vox10_1200.ply
Encoding time:  16.618353605270386
Models:  ['Model/train_2510/16_5_3/best_val_checkpoint_model__lr_50_b_2_da_0_nores_8_schedule_50.75_nobins_16_noltfil_5-epoch=01-val_loss=18.77.ckpt']
Occupied Voxels: 805285
Color bitstream:  Output/loot_vox10_1200/2510_pooling.color.bin
Metadata bitstream Output/loot_vox10_1200/2510_pooling.metadata.bin
Encoding information:  Output/loot_vox10_1200/2510_pooling.info.pkl
Metadata :  452
Bit per scale:  [ 0.74312426 11.70201734 44.81147035 42.74338805]
Average color bits per occupied voxels: 49.3366