-
Notifications
You must be signed in to change notification settings - Fork 16
/
run.sh
executable file
·233 lines (198 loc) · 10.2 KB
/
run.sh
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
export GPU="0"
export CAMERA="PINHOLE"
export EXP_NAME_1="stage1"
export EXP_NAME_2="stage2"
export EXP_NAME_3="stage3"
#
# Ensure that the following environment variables are accessible to the script:
# PROJECT_DIR and DATA_PATH
#
# Need to use this to activate conda environments
eval "$(conda shell.bash hook)"
#################
# PREPROCESSING #
#################
# Arrange raw images into a 3D Gaussian Splatting format
conda deactivate && conda activate gaussian_splatting_hair
cd $PROJECT_DIR/src/preprocessing
CUDA_VISIBLE_DEVICES="$GPU" python preprocess_raw_images.py \
--data_path $DATA_PATH
# Run COLMAP reconstruction and undistort the images and cameras
conda deactivate && conda activate gaussian_splatting_hair
cd $PROJECT_DIR/src
CUDA_VISIBLE_DEVICES="$GPU" python convert.py -s $DATA_PATH \
--camera $CAMERA --max_size 1024
# Run Matte-Anything
conda deactivate && conda activate matte_anything
cd $PROJECT_DIR/src/preprocessing
CUDA_VISIBLE_DEVICES="$GPU" python calc_masks.py \
--data_path $DATA_PATH --image_format png --max_size 2048
# Filter images using their IQA scores
conda deactivate && conda activate gaussian_splatting_hair
cd $PROJECT_DIR/src/preprocessing
CUDA_VISIBLE_DEVICES="$GPU" python filter_extra_images.py \
--data_path $DATA_PATH --max_imgs 128
# Resize images
conda deactivate && conda activate gaussian_splatting_hair
cd $PROJECT_DIR/src/preprocessing
CUDA_VISIBLE_DEVICES="$GPU" python resize_images.py --data_path $DATA_PATH
# Calculate orientation maps
conda deactivate && conda activate gaussian_splatting_hair
cd $PROJECT_DIR/src/preprocessing
CUDA_VISIBLE_DEVICES="$GPU" python calc_orientation_maps.py \
--img_path $DATA_PATH/images_2 \
--mask_path $DATA_PATH/masks_2/hair \
--orient_dir $DATA_PATH/orientations_2/angles \
--conf_dir $DATA_PATH/orientations_2/vars \
--filtered_img_dir $DATA_PATH/orientations_2/filtered_imgs \
--vis_img_dir $DATA_PATH/orientations_2/vis_imgs
# Run OpenPose
conda deactivate && cd $PROJECT_DIR/ext/openpose
mkdir $DATA_PATH/openpose
CUDA_VISIBLE_DEVICES="$GPU" ./build/examples/openpose/openpose.bin \
--image_dir $DATA_PATH/images_4 \
--scale_number 4 --scale_gap 0.25 --face --hand --display 0 \
--write_json $DATA_PATH/openpose/json \
--write_images $DATA_PATH/openpose/images --write_images_format jpg
# Run Face-Alignment
conda deactivate && conda activate gaussian_splatting_hair
cd $PROJECT_DIR/src/preprocessing
CUDA_VISIBLE_DEVICES="$GPU" python calc_face_alignment.py \
--data_path $DATA_PATH --image_dir "images_4"
# Run PIXIE
conda deactivate && conda activate pixie-env
cd $PROJECT_DIR/ext/PIXIE
CUDA_VISIBLE_DEVICES="$GPU" python demos/demo_fit_face.py \
-i $DATA_PATH/images_4 -s $DATA_PATH/pixie \
--saveParam True --lightTex False --useTex False \
--rasterizer_type pytorch3d
# Merge all PIXIE predictions in a single file
conda deactivate && conda activate gaussian_splatting_hair
cd $PROJECT_DIR/src/preprocessing
CUDA_VISIBLE_DEVICES="$GPU" python merge_smplx_predictions.py \
--data_path $DATA_PATH
# Convert COLMAP cameras to txt
conda deactivate && conda activate gaussian_splatting_hair
mkdir $DATA_PATH/sparse_txt
CUDA_VISIBLE_DEVICES="$GPU" colmap model_converter \
--input_path $DATA_PATH/sparse/0 \
--output_path $DATA_PATH/sparse_txt --output_type TXT
# Convert COLMAP cameras to H3DS format
conda deactivate && conda activate gaussian_splatting_hair
cd $PROJECT_DIR/src/preprocessing
CUDA_VISIBLE_DEVICES="$GPU" python colmap_parsing.py \
--path_to_scene $DATA_PATH
# Remove raw files to preserve disk space
rm -rf $DATA_PATH/input $DATA_PATH/images $DATA_PATH/masks $DATA_PATH/iqa*
##################
# RECONSTRUCTION #
##################
export EXP_PATH_1=$DATA_PATH/3d_gaussian_splatting/$EXP_NAME_1
# Run 3D Gaussian Splatting reconstruction
conda activate gaussian_splatting_hair && cd $PROJECT_DIR/src
CUDA_VISIBLE_DEVICES="$GPU" python train_gaussians.py \
-s $DATA_PATH -m "$EXP_PATH_1" -r 1 --port "888$GPU" \
--trainable_cameras --trainable_intrinsics --use_barf \
--lambda_dorient 0.1
# Run FLAME mesh fitting
conda activate gaussian_splatting_hair
cd $PROJECT_DIR/ext/NeuralHaircut/src/multiview_optimization
CUDA_VISIBLE_DEVICES="$GPU" python fit.py --conf confs/train_person_1.conf \
--batch_size 1 --train_rotation True --fixed_images True \
--save_path $DATA_PATH/flame_fitting/$EXP_NAME_1/stage_1 \
--data_path $DATA_PATH \
--fitted_camera_path $EXP_PATH_1/cameras/30000_matrices.pkl
CUDA_VISIBLE_DEVICES="$GPU" python fit.py --conf confs/train_person_1.conf \
--batch_size 4 --train_rotation True --fixed_images True \
--save_path $DATA_PATH/flame_fitting/$EXP_NAME_1/stage_2 \
--checkpoint_path $DATA_PATH/flame_fitting/$EXP_NAME_1/stage_1/opt_params_final \
--data_path $DATA_PATH \
--fitted_camera_path $EXP_PATH_1/cameras/30000_matrices.pkl
CUDA_VISIBLE_DEVICES="$GPU" python fit.py --conf confs/train_person_1_.conf \
--batch_size 32 --train_rotation True --train_shape True \
--save_path $DATA_PATH/flame_fitting/$EXP_NAME_1/stage_3 \
--checkpoint_path $DATA_PATH/flame_fitting/$EXP_NAME_1/stage_2/opt_params_final \
--data_path $DATA_PATH \
--fitted_camera_path $EXP_PATH_1/cameras/30000_matrices.pkl
# Crop the reconstructed scene
conda activate gaussian_splatting_hair && cd $PROJECT_DIR/src/preprocessing
CUDA_VISIBLE_DEVICES="$GPU" python scale_scene_into_sphere.py \
--path_to_data $DATA_PATH \
-m "$DATA_PATH/3d_gaussian_splatting/$EXP_NAME_1" --iter 30000
# Remove hair Gaussians that intersect with the FLAME head mesh
conda activate gaussian_splatting_hair && cd $PROJECT_DIR/src/preprocessing
CUDA_VISIBLE_DEVICES="$GPU" python filter_flame_intersections.py \
--flame_mesh_dir $DATA_PATH/flame_fitting/$EXP_NAME_1 \
-m "$DATA_PATH/3d_gaussian_splatting/$EXP_NAME_1" --iter 30000 \
--project_dir $PROJECT_DIR/ext/NeuralHaircut
# Run rendering for training views
conda activate gaussian_splatting_hair && cd $PROJECT_DIR/src
CUDA_VISIBLE_DEVICES="$GPU" python render_gaussians.py \
-s $DATA_PATH -m "$DATA_PATH/3d_gaussian_splatting/$EXP_NAME_1" \
--skip_test --scene_suffix "_cropped" --iteration 30000 \
--trainable_cameras --trainable_intrinsics --use_barf
# Get FLAME mesh scalp maps
conda activate gaussian_splatting_hair && cd $PROJECT_DIR/src/preprocessing
CUDA_VISIBLE_DEVICES="$GPU" python extract_non_visible_head_scalp.py \
--project_dir $PROJECT_DIR/ext/NeuralHaircut --data_dir $DATA_PATH \
--flame_mesh_dir $DATA_PATH/flame_fitting/$EXP_NAME_1 \
--cams_path $DATA_PATH/3d_gaussian_splatting/$EXP_NAME_1/cameras/30000_matrices.pkl \
-m "$DATA_PATH/3d_gaussian_splatting/$EXP_NAME_1"
# Run latent hair strands reconstruction
conda activate gaussian_splatting_hair && cd $PROJECT_DIR/src
CUDA_VISIBLE_DEVICES="$GPU" python train_latent_strands.py \
-s $DATA_PATH -m "$DATA_PATH/3d_gaussian_splatting/$EXP_NAME_1" -r 1 \
--model_path_hair "$DATA_PATH/strands_reconstruction/$EXP_NAME_2" \
--flame_mesh_dir "$DATA_PATH/flame_fitting/$EXP_NAME_1" \
--pointcloud_path_head "$EXP_PATH_1/point_cloud_filtered/iteration_30000/raw_point_cloud.ply" \
--hair_conf_path "$PROJECT_DIR/src/arguments/hair_strands_textured.yaml" \
--lambda_dmask 0.1 --lambda_dorient 0.1 --lambda_dsds 0.01 \
--load_synthetic_rgba --load_synthetic_geom --binarize_masks --iteration_data 30000 \
--trainable_cameras --trainable_intrinsics --use_barf \
--iterations 20000 --port "800$GPU"
# Run hair strands reconstruction
conda activate gaussian_splatting_hair && cd $PROJECT_DIR/src
CUDA_VISIBLE_DEVICES="$GPU" python train_strands.py \
-s $DATA_PATH -m "$DATA_PATH/3d_gaussian_splatting/$EXP_NAME_1" -r 1 \
--model_path_curves "$DATA_PATH/curves_reconstruction/$EXP_NAME_3" \
--flame_mesh_dir "$DATA_PATH/flame_fitting/$EXP_NAME_1" \
--pointcloud_path_head "$EXP_PATH_1/point_cloud_filtered/iteration_30000/raw_point_cloud.ply" \
--start_checkpoint_hair "$DATA_PATH/strands_reconstruction/$EXP_NAME_2/checkpoints/20000.pth" \
--hair_conf_path "$PROJECT_DIR/src/arguments/hair_strands_textured.yaml" \
--lambda_dmask 0.1 --lambda_dorient 0.1 --lambda_dsds 0.01 \
--load_synthetic_rgba --load_synthetic_geom --binarize_masks --iteration_data 30000 \
--position_lr_init 0.0000016 --position_lr_max_steps 10000 \
--trainable_cameras --trainable_intrinsics --use_barf \
--iterations 10000 --port "800$GPU"
rm -rf "$DATA_PATH/3d_gaussian_splatting/$EXP_NAME_1/train_cropped"
##################
# VISUALIZATIONS #
##################
# Export the resulting strands as pkl and ply
conda activate gaussian_splatting_hair && cd $PROJECT_DIR/src/preprocessing
CUDA_VISIBLE_DEVICES="$GPU" python export_curves.py \
--data_dir $DATA_PATH --model_name $EXP_NAME_3 --iter 10000 \
--flame_mesh_path "$DATA_PATH/flame_fitting/$EXP_NAME_1/stage_3/mesh_final.obj" \
--scalp_mesh_path "$DATA_PATH/flame_fitting/$EXP_NAME_1/scalp_data/scalp.obj" \
--hair_conf_path "$PROJECT_DIR/src/arguments/hair_strands_textured.yaml"
# Render the visualizations
conda activate gaussian_splatting_hair && cd $PROJECT_DIR/src/postprocessing
CUDA_VISIBLE_DEVICES="$GPU" python render_video.py \
--blender_path "$BLENDER_DIR" --input_path "$DATA_PATH" \
--exp_name_1 "$EXP_NAME_1" --exp_name_3 "$EXP_NAME_3"
# Render the strands
conda activate gaussian_splatting_hair && cd $PROJECT_DIR/src
CUDA_VISIBLE_DEVICES="$GPU" python render_strands.py \
-s $DATA_PATH --data_dir "$DATA_PATH" --data_device 'cpu' --skip_test \
-m "$DATA_PATH/3d_gaussian_splatting/$EXP_NAME_1" --iteration 30000 \
--flame_mesh_dir "$DATA_PATH/flame_fitting/$EXP_NAME_1" \
--model_hair_path "$DATA_PATH/curves_reconstruction/$EXP_NAME_3" \
--hair_conf_path "$PROJECT_DIR/src/arguments/hair_strands_textured.yaml" \
--checkpoint_hair "$DATA_PATH/strands_reconstruction/$EXP_NAME_2/checkpoints/20000.pth" \
--checkpoint_curves "$DATA_PATH/curves_reconstruction/$EXP_NAME_3/checkpoints/10000.pth" \
--pointcloud_path_head "$EXP_PATH_1/point_cloud/iteration_30000/raw_point_cloud.ply" \
--interpolate_cameras
# Make the video
conda activate gaussian_splatting_hair && cd $PROJECT_DIR/src/postprocessing
CUDA_VISIBLE_DEVICES="$GPU" python concat_video.py \
--input_path "$DATA_PATH" --exp_name_3 "$EXP_NAME_3"