NJJ - Bas Relief - Hotel Lambert

[rebecca-lay3rs]

Data treatment

• cut videos
• rename frames from videos so it is not duplicated for i in *.png; do mv -- "$i" "${i%.png}_7037.png"; done
• save all frames to one folder

First results using 3DGS and Sugar

• with -r "sdf" --low_poly True --poisson_depth 8

Videos Xavier

basrelief2

• [x] camera poses COLMAP RTX

COLMAP	Alignement

• [x] training 3DGS H100 (7k and 30k iters)
• [x] training SuGaR

basrelief4

• [x] camera poses COLMAP

COLMAP	Alignement

• [x] training 3DGS H100 (7k and 30k iters)
• [x] training SuGaR

python train.py -s data/basrelief4/ -c outputs_GS/basrelief4/ -r "sdf" --low_poly True --poisson_depth 8 --experiment_name basrelief4

result:

• not enough view points on the right
• mesh with holes

3DGS Rendering	Mesh

basrelief6

• [x] camera poses COLMAP

COLMAP	Alignement

• [x] training 3DGS H100 (7k and 30k iters)
• [x] training SuGaR

result:

3DGS Rendering	Mesh

basrelief8

• [x] camera poses COLMAP

COLMAP	Alignement

• [x] training 3DGS H100 (7k and 30k iters)
• [x] training SuGaR

result:

3DGS Rendering	Mesh

basrelief10

• [x] camera poses COLMAP

COLMAP	Alignement

• [x] training 3DGS H100 (7k and 30k iters)
• [x] training SuGaR

result:

3DGS Rendering	Mesh

extra:

basrelief12

• [x] camera poses COLMAP

COLMAP	Alignement

• [x] training 3DGS H100 (7k and 30k iters)
• [x] training SuGaR

result:

3DGS Rendering	Mesh

extra:

Videos Victoriya

b3

• [x] camera poses PPR

PPR	Alignement

• [x] training 3DGS H100 (7k and 30k iters)
• [x] training SuGaR -> check mesh

python train.py -s data/b3/ -c outputs_GS/b3/ -r "sdf" --low_poly True --poisson_depth 8 --experiment_name b3

result:

• a lot of floaters when parallel to the object (missing parallel view points)
• many holes in the mesh

3DGS Rendering	Mesh

from the sides:

Template:	3DGS Rendering	Mesh

[rebecca-lay3rs]

Trying to improve the best result: basrelief10

--vertices_density_quantile 0.2 --poisson_depth 9 -l 0.3

DATA_PATH=data/basrelief10/
GS_PATH=outputs_GS/basrelief10/ 
COARSE_MODEL_PATH=output/coarse/basrelief10_vdq0_ncf0p2_pd8_sdf/sugarcoarse_3Dgs7000_sdfestim05_sdfnorm02/15000.pt
EXPERIMENT_NAME=basrelief10_vdq0p2_ncf0p2_pd8_sdf/
COARSE_MESH_PATH=output/coarse_mesh/${EXPERIMENT_NAME}

python extract_mesh.py -s ${DATA_PATH} -c ${GS_PATH} -i 7000 -m ${COARSE_MODEL_PATH} -l 0.3 -d 200000 -o ${COARSE_MESH_PATH} --experiment_name ${EXPERIMENT_NAME} --normal_consistency_factor 0.2 --vertices_density_quantile 0.2 --poisson_depth 9

--vertices_density_quantile 0.2 --poisson_depth 8 -l 0.3

DATA_PATH=data/basrelief10/
GS_PATH=outputs_GS/basrelief10/ 
COARSE_MODEL_PATH=output/coarse/basrelief10_vdq0_ncf0p2_pd8_sdf/sugarcoarse_3Dgs7000_sdfestim05_sdfnorm02/15000.pt
EXPERIMENT_NAME=basrelief10_vdq0p2_ncf0p2_pd8_sdf/
COARSE_MESH_PATH=output/coarse_mesh/${EXPERIMENT_NAME}

python extract_mesh.py -s ${DATA_PATH} -c ${GS_PATH} -i 7000 -m ${COARSE_MODEL_PATH} -l 0.3 -d 200000 -o ${COARSE_MESH_PATH} --experiment_name ${EXPERIMENT_NAME} --normal_consistency_factor 0.2 --vertices_density_quantile 0.2 --poisson_depth 8

--vertices_density_quantile 0.0 --poisson_depth 7 -l 0.3

DATA_PATH=data/basrelief10/
GS_PATH=outputs_GS/basrelief10/ 
COARSE_MODEL_PATH=output/coarse/basrelief10_vdq0_ncf0p2_pd8_sdf/sugarcoarse_3Dgs7000_sdfestim05_sdfnorm02/15000.pt
EXPERIMENT_NAME=basrelief10_vdq0_ncf0p2_pd7_sdf/
COARSE_MESH_PATH=output/coarse_mesh/${EXPERIMENT_NAME}

python extract_mesh.py -s ${DATA_PATH} -c ${GS_PATH} -i 7000 -m ${COARSE_MODEL_PATH} -l 0.3 -d 200000 -o ${COARSE_MESH_PATH} --experiment_name ${EXPERIMENT_NAME} --normal_consistency_factor 0.2 --vertices_density_quantile 0.0 --poisson_depth 7

--vertices_density_quantile 0.0 --poisson_depth 7 -l 0.3

DATA_PATH=data/basrelief10/ GS_PATH=outputs_GS/basrelief10/ COARSE_MODEL_PATH=output/coarse/basrelief10_vdq0_ncf0p2_pd8_sdf/sugarcoarse_3Dgs7000_sdfestim05_sdfnorm02/15000.pt EXPERIMENT_NAME=basrelief10_vdq0_ncf0p2_pd7_sdf_d1M/ COARSE_MESH_PATH=output/coarse_mesh/${EXPERIMENT_NAME}

python extract_mesh.py -s ${DATA_PATH} -c ${GS_PATH} -i 7000 -m ${COARSE_MODEL_PATH} -l 0.3 -d 1000000 -o ${COARSE_MESH_PATH} --experiment_name ${EXPERIMENT_NAME} --normal_consistency_factor 0.2 --vertices_density_quantile 0.0 --poisson_depth 7

--vertices_density_quantile 0.0 --poisson_depth 8 -l 0.1

DATA_PATH=data/basrelief10/ GS_PATH=outputs_GS/basrelief10/ COARSE_MODEL_PATH=output/coarse/basrelief10_vdq0_ncf0p2_pd8_sdf/sugarcoarse_3Dgs7000_sdfestim05_sdfnorm02/15000.pt EXPERIMENT_NAME=basrelief10_vdq0_ncf0p2_pd8_sdf_l0p1/ COARSE_MESH_PATH=output/coarse_mesh/${EXPERIMENT_NAME}

python extract_mesh.py -s ${DATA_PATH} -c ${GS_PATH} -i 7000 -m ${COARSE_MODEL_PATH} -l 0.1 -d 200000 -o ${COARSE_MESH_PATH} --experiment_name ${EXPERIMENT_NAME} --normal_consistency_factor 0.2 --vertices_density_quantile 0.0 --poisson_depth 8

--vertices_density_quantile 0.0 --poisson_depth 8 -l 0.5

DATA_PATH=data/basrelief10/ GS_PATH=outputs_GS/basrelief10/ COARSE_MODEL_PATH=output/coarse/basrelief10_vdq0_ncf0p2_pd8_sdf/sugarcoarse_3Dgs7000_sdfestim05_sdfnorm02/15000.pt EXPERIMENT_NAME=basrelief10_vdq0_ncf0p2_pd8_sdf_l0p5/ COARSE_MESH_PATH=output/coarse_mesh/${EXPERIMENT_NAME}

python extract_mesh.py -s ${DATA_PATH} -c ${GS_PATH} -i 7000 -m ${COARSE_MODEL_PATH} -l 0.5 -d 200000 -o ${COARSE_MESH_PATH} --experiment_name ${EXPERIMENT_NAME} --normal_consistency_factor 0.2 --vertices_density_quantile 0.0 --poisson_depth 8

--vertices_density_quantile 0.0 --poisson_depth 8 -l 0.5 --normal_consistency_factor 0.4

python train.py -s data/basrelief10/ -c outputs_GS/basrelief10/ -r "sdf" --low_poly True --poisson_depth 8 --normal_consistency_factor 0.4 --vertices_density_quantile 0. --experiment_name basrelief10_vdq0_ncf0p4_pd8_sdf

retrain the sugar coarse model using "density"

python train.py -s data/basrelief10/ -c outputs_GS/basrelief10/ -r "density" --low_poly True --poisson_depth 8 --normal_consistency_factor 0.2 --vertices_density_quantile 0. --experiment_name basrelief10_vdq0_ncf0p2_pd8_sdf