search

ewrfcas / MVSFormer

Codes of MVSFormer: Multi-View Stereo by Learning Robust Image Features and Temperature-based Depth (TMLR2023)
Apache License 2.0
175 stars 10 forks source link

Lots of noise in output ply in tnt dataset #20

Closed CC9310 closed 1 year ago

CC9310 commented 1 year ago

Thanks for the author's great work. Following the tnt test instruction, Lots of noise in the output ply 2023-06-06_195304 bash: CUDA_VISIBLE_DEVICES=0 python test.py --dataset tt --batch_size 1 --testpath /data/tankandtemples/intermediate/ \ --testlist ./lists/tanksandtemples/intermediate.txt \ --resume "/home/methods/MVSFormer/pretrained_weights/MVSFormer/best.pth" \ --outdir /data4/trained_models/MVSFormer_tnt_nview10/ \ --interval_scale 1.0 --num_view 10 --numdepth 256 --max_h 1088 --max_w 1920 \ --filter_method dpcd --prob_threshold 0.5,0.5,0.5,0.5 \ --use_short_range --combine_conf --tmps 5.0,5.0,5.0,1.0 Do the author has any idea to solve the problem?

ewrfcas commented 1 year ago

Hi, we recommend using MVSFormer-blended to evaluate Tanks and Temples.

CC9310 commented 1 year ago

Thanks for quick reply :-). While, after using MVSFormer-Blended/best.pth with src_view 10, the recons point cloud also very noise(I didn't modify any parameter in the code).

ewrfcas commented 1 year ago

Sorry for the confusing code. We used --conf for T&T and --prob_threshold for DTU. I will unify them latter.

ewrfcas commented 1 year ago

Codes are updated.

CC9310 commented 1 year ago

Codes are updated.

it's very kind of you :-)