first commit

scripts/install_ext.sh

@@ -0,0 +1,4 @@
+pip install ./raymarching
+pip install ./shencoder
+pip install ./freqencoder
+pip install ./gridencoder

scripts/magic123/preprocess_folder.sh

@@ -0,0 +1,6 @@
+topdir=$1
+imagename=$2 # rgba.png or image.png
+for i in $topdir/*; do
+    echo preprocessing "$i"/$imagename ...
+    python scripts/preprocess_image.py "$i"/$imagename
+done

scripts/magic123/run_2dprior.sh

@@ -0,0 +1,75 @@
+#! /bin/bash
+#SBATCH -N 1
+#SBATCH --array=0
+#SBATCH -J magic123
+#SBATCH -o slurm_logs/%x.%3a.%A.out
+#SBATCH -e slurm_logs/%x.%3a.%A.err
+#SBATCH --time=3:00:00
+#SBATCH --gres=gpu:v100:1
+#SBATCH --cpus-per-gpu=6
+#SBATCH --mem=30G
+##SBATCH --gpus=1
+
+module load gcc/7.5.0
+
+
+#source ~/.bashrc
+#source activate magic123
+source venv_magic123/bin/activate
+which python 
+
+nvidia-smi
+nvcc --version
+
+hostname
+NUM_GPU_AVAILABLE=`nvidia-smi --query-gpu=name --format=csv,noheader | wc -l`
+echo "number of gpus:" $NUM_GPU_AVAILABLE
+
+RUN_ID=$2
+RUN_ID2=$3
+DATA_DIR=$4
+IMAGE_NAME=$5
+step1=$6
+step2=$7
+FILENAME=$(basename $DATA_DIR)
+dataset=$(basename $(dirname $DATA_DIR))
+echo reconstruct $FILENAME under dataset $dataset from folder $DATA_DIR ...
+
+if (( ${step1} )); then
+    CUDA_VISIBLE_DEVICES=$1 python main.py -O \
+        --text "A high-resolution DSLR image of <token>" \
+        --sd_version 1.5 \
+        --image ${DATA_DIR}/${IMAGE_NAME} \
+        --learned_embeds_path ${DATA_DIR}/learned_embeds.bin \
+        --workspace out/magic123-2d/magic123-2d-${RUN_ID}-coarse/$dataset/magic123_2d_${FILENAME}_${RUN_ID}_coarse \
+        --optim adam \
+        --iters 5000 \
+        --guidance SD \
+        --lambda_guidance 1 \
+        --guidance_scale 100 \
+        --latent_iter_ratio 0 \
+        --normal_iter_ratio 0.2 \
+        --t_range 0.2 0.6 \
+        --bg_radius -1 \
+        --save_mesh \
+        ${@:8}
+fi
+
+if (( ${step2} )); then
+    CUDA_VISIBLE_DEVICES=$1 python main.py -O \
+        --text "A high-resolution DSLR image of <token>" \
+        --sd_version 1.5 \
+        --image ${DATA_DIR}/${IMAGE_NAME} \
+        --learned_embeds_path ${DATA_DIR}/learned_embeds.bin \
+        --workspace out/magic123-2d/magic123-2d-${RUN_ID}-${RUN_ID2}/$dataset/magic123_2d_${FILENAME}_${RUN_ID}_${RUN_ID2} \
+        --dmtet --init_ckpt out/magic123-2d/magic123-2d-${RUN_ID}-coarse/$dataset/magic123_2d_${FILENAME}_${RUN_ID}_coarse/checkpoints/magic123_2d_${FILENAME}_${RUN_ID}_coarse.pth \
+        --iters 5000 \
+        --optim adam \
+        --latent_iter_ratio 0 \
+        --guidance SD \
+        --lambda_guidance 1e-3 \
+        --guidance_scale 100 \
+        --rm_edge \
+        --bg_radius -1 \
+        --save_mesh 
+fi

scripts/magic123/run_2dprior_notextinv_ironman.sh

@@ -0,0 +1,73 @@
+#! /bin/bash
+#SBATCH -N 1
+#SBATCH --array=0
+#SBATCH -J magic123
+#SBATCH -o slurm_logs/%x.%3a.%A.out
+#SBATCH -e slurm_logs/%x.%3a.%A.err
+#SBATCH --time=3:00:00
+#SBATCH --gres=gpu:v100:1
+#SBATCH --cpus-per-gpu=6
+#SBATCH --mem=30G
+##SBATCH --gpus=1
+
+module load gcc/7.5.0
+
+
+#source ~/.bashrc
+#source activate magic123
+source venv_magic123/bin/activate
+which python 
+
+nvidia-smi
+nvcc --version
+
+hostname
+NUM_GPU_AVAILABLE=`nvidia-smi --query-gpu=name --format=csv,noheader | wc -l`
+echo "number of gpus:" $NUM_GPU_AVAILABLE
+
+RUN_ID=$2
+RUN_ID2=$3
+DATA_DIR="data/demo/ironman"
+IMAGE_NAME="rgba.png"
+step1=$4
+step2=$5
+FILENAME=$(basename $DATA_DIR)
+dataset=$(basename $(dirname $DATA_DIR))
+echo reconstruct $FILENAME under dataset $dataset from folder $DATA_DIR ...
+
+if (( ${step1} )); then
+    CUDA_VISIBLE_DEVICES=$1 python main.py -O \
+        --text "A high-resolution DSLR image of a full body ironman" \
+        --sd_version 1.5 \
+        --image ${DATA_DIR}/${IMAGE_NAME} \
+        --workspace out/magic123-2d-noinv/magic123-2d-noinv-${RUN_ID}-coarse/$dataset/magic123-2d-noinv_${FILENAME}_${RUN_ID}_coarse \
+        --optim adam \
+        --iters 5000 \
+        --guidance SD \
+        --lambda_guidance 1 \
+        --guidance_scale 100 \
+        --latent_iter_ratio 0 \
+        --normal_iter_ratio 0.2 \
+        --t_range 0.2 0.6 \
+        --bg_radius -1 \
+        --save_mesh \
+        ${@:6}
+fi
+
+if (( ${step2} )); then
+    CUDA_VISIBLE_DEVICES=$1 python main.py -O \
+        --text "A high-resolution DSLR image of a full body ironman" \
+        --sd_version 1.5 \
+        --image ${DATA_DIR}/${IMAGE_NAME} \
+        --workspace out/magic123-2d-noinv/magic123-2d-noinv-${RUN_ID}-${RUN_ID2}/$dataset/magic123-2d-noinv_${FILENAME}_${RUN_ID}_${RUN_ID2} \
+        --dmtet --init_ckpt out/magic123-2d-noinv/magic123-2d-noinv-${RUN_ID}-coarse/$dataset/magic123-2d-noinv_${FILENAME}_${RUN_ID}_coarse/checkpoints/magic123-2d-noinv_${FILENAME}_${RUN_ID}_coarse.pth \
+        --iters 5000 \
+        --optim adam \
+        --latent_iter_ratio 0 \
+        --guidance SD \
+        --lambda_guidance 1e-3 \
+        --guidance_scale 100 \
+        --rm_edge \
+        --bg_radius -1 \
+        --save_mesh 
+fi

scripts/magic123/run_3dprior.sh

@@ -0,0 +1,73 @@
+#! /bin/bash
+#SBATCH -N 1
+#SBATCH --array=0
+#SBATCH -J magic123
+#SBATCH -o slurm_logs/%x.%3a.%A.out
+#SBATCH -e slurm_logs/%x.%3a.%A.err
+#SBATCH --time=3:00:00
+#SBATCH --gres=gpu:v100:1
+#SBATCH --cpus-per-gpu=6
+#SBATCH --mem=30G
+##SBATCH --gpus=1
+
+module load gcc/7.5.0
+
+
+#source ~/.bashrc
+#source activate magic123
+source venv_magic123/bin/activate
+which python 
+
+nvidia-smi
+nvcc --version
+
+hostname
+NUM_GPU_AVAILABLE=`nvidia-smi --query-gpu=name --format=csv,noheader | wc -l`
+echo "number of gpus:" $NUM_GPU_AVAILABLE
+
+RUN_ID=$2
+RUN_ID2=$3
+DATA_DIR=$4
+IMAGE_NAME=$5
+step1=$6
+step2=$7
+FILENAME=$(basename $DATA_DIR)
+dataset=$(basename $(dirname $DATA_DIR))
+echo reconstruct $FILENAME under dataset $dataset from folder $DATA_DIR ...
+
+if (( ${step1} )); then
+    CUDA_VISIBLE_DEVICES=$1 python main.py -O \
+        --text "A high-resolution DSLR image" \
+        --sd_version 1.5 \
+        --image ${DATA_DIR}/${IMAGE_NAME} \
+        --workspace out/magic123-3d/magic123-3d-${RUN_ID}/$dataset/magic123-3d_${FILENAME}_${RUN_ID}_5k \
+        --optim adam \
+        --iters 5000 \
+        --guidance zero123 \
+        --lambda_guidance 40 \
+        --guidance_scale 5 \
+        --latent_iter_ratio 0 \
+        --normal_iter_ratio 0.2 \
+        --t_range 0.2 0.6 \
+        --bg_radius -1 \
+        --save_mesh \
+        ${@:8}
+fi
+
+if (( ${step2} )); then
+    CUDA_VISIBLE_DEVICES=$1 python main.py -O \
+        --text "A high-resolution DSLR image" \
+        --sd_version 1.5 \
+        --image ${DATA_DIR}/${IMAGE_NAME} \
+        --workspace out/magic123-3d/magic123-3d-${RUN_ID}-${RUN_ID2}/$dataset/magic123-3d_${FILENAME}_${RUN_ID}_${RUN_ID2} \
+        --dmtet --init_ckpt out/magic123-3d/magic123-3d-${RUN_ID}/$dataset/magic123-3d_${FILENAME}_${RUN_ID}_5k/checkpoints/magic123-3d_${FILENAME}_${RUN_ID}_5k.pth \
+        --iters 5000 \
+        --optim adam \
+        --latent_iter_ratio 0 \
+        --guidance zero123 \
+        --lambda_guidance 0.01 \
+        --guidance_scale 5 \
+        --bg_radius -1 \
+        --rm_edge \
+        --save_mesh 
+fi

scripts/magic123/run_both_priors.sh

@@ -0,0 +1,76 @@
+#! /bin/bash
+#SBATCH -N 1
+#SBATCH --array=0
+#SBATCH -J magic123
+#SBATCH -o slurm_logs/%x.%3a.%A.out
+#SBATCH -e slurm_logs/%x.%3a.%A.err
+#SBATCH --time=3:00:00
+#SBATCH --gres=gpu:v100:1
+#SBATCH --cpus-per-gpu=6
+#SBATCH --mem=30G
+##SBATCH --gpus=1
+
+module load gcc/7.5.0
+
+
+#source ~/.bashrc
+#source activate magic123
+source venv_magic123/bin/activate
+which python 
+
+nvidia-smi
+nvcc --version
+
+hostname
+NUM_GPU_AVAILABLE=`nvidia-smi --query-gpu=name --format=csv,noheader | wc -l`
+echo "number of gpus:" $NUM_GPU_AVAILABLE
+
+RUN_ID=$2 # jobname for the first stage
+RUN_ID2=$3 # jobname for the second stage
+DATA_DIR=$4 # path to the directory containing the images, e.g. data/nerf4/chair
+IMAGE_NAME=$5 # name of the image file, e.g. rgba.png
+step1=$6 # whether to use the first stage
+step2=$7 # whether to use the second stage
+
+FILENAME=$(basename $DATA_DIR)
+dataset=$(basename $(dirname $DATA_DIR))
+echo reconstruct $FILENAME under dataset $dataset from folder $DATA_DIR ...
+
+if (( ${step1} )); then
+    CUDA_VISIBLE_DEVICES=$1 python main.py -O \
+        --text "A high-resolution DSLR image of <token>" \
+        --sd_version 1.5 \
+        --image ${DATA_DIR}/${IMAGE_NAME} \
+        --learned_embeds_path ${DATA_DIR}/learned_embeds.bin \
+        --workspace out/magic123-${RUN_ID}-coarse/$dataset/magic123_${FILENAME}_${RUN_ID}_coarse \
+        --optim adam \
+        --iters 5000 \
+        --guidance SD zero123 \
+        --lambda_guidance 1.0 40 \
+        --guidance_scale 100 5 \
+        --latent_iter_ratio 0 \
+        --normal_iter_ratio 0.2 \
+        --t_range 0.2 0.6 \
+        --bg_radius -1 \
+        --save_mesh \
+        ${@:8}
+fi
+
+if (( ${step2} )); then
+    CUDA_VISIBLE_DEVICES=$1 python main.py -O \
+        --text "A high-resolution DSLR image of <token>" \
+        --sd_version 1.5 \
+        --image ${DATA_DIR}/${IMAGE_NAME} \
+        --learned_embeds_path ${DATA_DIR}/learned_embeds.bin \
+        --workspace out/magic123-${RUN_ID}-${RUN_ID2}/$dataset/magic123_${FILENAME}_${RUN_ID}_${RUN_ID2} \
+        --dmtet --init_ckpt out/magic123-${RUN_ID}-coarse/$dataset/magic123_${FILENAME}_${RUN_ID}_coarse/checkpoints/magic123_${FILENAME}_${RUN_ID}_coarse.pth \
+        --iters 5000 \
+        --optim adam \
+        --latent_iter_ratio 0 \
+        --guidance SD zero123 \
+        --lambda_guidance 1e-3 0.01 \
+        --guidance_scale 100 5 \
+        --rm_edge \
+        --bg_radius -1 \
+        --save_mesh 
+fi

scripts/magic123/run_folder_both_priors.sh

@@ -0,0 +1,13 @@
+device=$1
+runid=$2 # jobname for the first stage
+runid2=$3 # jobname for the second stage
+topdir=$4   # path to the directory containing the images, e.g. data/nerf4
+imagename=$5
+step1=$6
+step2=$7
+
+for i in $topdir/*; do
+    echo "$i"
+    [ -d "$i" ] && echo "$i exists."
+    bash scripts/magic123/run_both_priors.sh $device $runid "$i" $imagename $step1 $step2 ${@:8}
+done

scripts/magic123/run_list_both_priors.sh

@@ -0,0 +1,25 @@
+device=$1
+runid=$2 # jobname for the first stage
+runid2=$3 # jobname for the second stage
+imagename=$4
+step1=$5
+step2=$6
+
+examples=(
+    'data/realfusion15/teddy_bear/'
+    'data/realfusion15/mental_dragon_statue/'
+    'data/realfusion15/colorful_teapot/'
+    'data/realfusion15/fish_real_nemo/'
+    'data/realfusion15/two_cherries' 
+    'data/realfusion15/watercolor_horse/'
+    'data/nerf4/chair'
+    'data/nerf4/drums'
+    'data/nerf4/ficus'
+    'data/nerf4/mic'
+)
+
+for i in "${examples[@]}"; do
+    echo "$i"
+    [ -d "$i" ] && echo "$i exists."
+    bash scripts/magic123/run_both_priors.sh $device $runid "$i" $imagename $step1 $step2 ${@:7}
+done

scripts/snap/aws_folder.sh

@@ -0,0 +1,21 @@
+script_name=$1
+runid=$2
+runid2=$3
+topdir=$4
+imagename=$5
+run1=$6
+run2=$7
+arguments="${@:8}"
+
+timestamp=$(date +'%Y%m%d')
+for i in $topdir/*; do
+    echo "$i"
+    [ -d "$i" ] && echo "$i exists."
+    example=$(basename $i)
+    echo ${@:8}
+    python scripts/snap/submit_cluster_job.py --yaml_folder scripts/snap/yamls \
+        --gpu_memory 40 --gpu_num 1 --force_node --cpu_num_per_gpu 6 --memory_per_gpu 30.0 --replicas 1 \
+        --project_name magic123 --project_support_alias img2mesh \
+        --job_name gqian-$timestamp-$runid-$runid2-$example \
+        --command "bash $script_name 0 $runid $runid2 $i $imagename $run1 $run2 $arguments "
+done

scripts/snap/aws_list.sh

@@ -0,0 +1,27 @@
+script_name=$1
+runid=$2
+runid2=$3
+imagename=$4
+run1=$5
+run2=$6
+arguments="${@:7}"
+
+
+examples=(
+    'data/realfusion15/two_donuts/'
+    'data/realfusion15/watercolor_horse/'
+)
+
+
+timestamp=$(date +'%Y%m%d')
+for i in "${examples[@]}"; do
+    echo "$i"
+    [ -d "$i" ] && echo "$i exists."
+    example=$(basename $i)
+    echo ${@:8}
+    python scripts/snap/submit_cluster_job.py --yaml_folder scripts/snap/yamls \
+        --gpu_memory 40 --gpu_num 1 --force_node --cpu_num_per_gpu 6 --memory_per_gpu 30.0 --replicas 1 \
+        --project_name magic123 --project_support_alias img2mesh \
+        --job_name gqian-$timestamp-$runid-$runid2-$example \
+        --command "bash $script_name 0 $runid $runid2 $i $imagename $run1 $run2 $arguments "
+done

scripts/snap/submit_cluster_debug.sh

@@ -0,0 +1,8 @@
+jobname=$1
+timestamp=$(date +'%Y%m%d')
+[ -d "$i" ] && echo "$i exists."
+python scripts/snap/submit_cluster_job.py --yaml_folder scripts/snap/yamls \
+        --gpu_memory 40 --gpu_num 1 --force_node --cpu_num_per_gpu 6 --memory_per_gpu 30.0 --replicas 1 \
+        --project_name magic123 --project_support_alias img2mesh \
+        --job_name gqian-$timestamp-$1 \
+        --command "while :; do sleep 1000; done"

scripts/snap/submit_cluster_job.py

@@ -0,0 +1,99 @@
+
+import yaml
+import os
+import argparse
+
+
+def generate_yaml(
+                  gpu_memory=40, # 40G or 80G?
+                  gpu_num=1, 
+                  cpu_num_per_gpu=6,
+                  memory_per_gpu=30,
+                  replicas=1,
+                  project_name='magic123',
+                  project_support_alias='img2mesh',
+                  pre_run_event='mkdir -p /fsx/code && ln -s /nfs/code/gqian /fsx/code/ && cd /fsx/code/gqian/img2mesh', 
+                  command="while :; do sleep 1000; done", 
+                  job_name='debug',  
+                  force_node=False,
+                  **kwargs
+                  ):
+    data = {
+        'docker_image': '440036398022.dkr.ecr.us-west-2.amazonaws.com/facecraft-ml:efa',
+        'project_name': project_name,
+        'project_support_alias': project_support_alias,
+        'team': 'creative_vision',
+        #'fsx': 'fs-0b933bba2f17fe699', # 100T genai filesystem
+        'fsx': 'fs-056caaa56fa5cc5f3', # 2T personal filesystem of gqian
+        'gpu_type': 'nvidia-tesla-a100',
+        'gpu_num': gpu_num,
+        'cpu_num': str(int(cpu_num_per_gpu * gpu_num)),
+        'memory': str(int(memory_per_gpu * gpu_num)),
+        'gpu_memory': str(int(gpu_memory)),
+        'pytorchjob': {
+            'replicas': replicas
+        },
+        'efa': True,
+        'script': {
+            'pre_run_event': str(pre_run_event),
+            'command': str(command),
+            'jobs': [
+                {'name': job_name}
+            ]
+        }
+    }
+    if gpu_num == 1 and not force_node:
+        gpu_yaml = {
+            'custom_node_labels': {
+            'use_case': 'p4d_debug',
+        },
+        }
+    else:
+        gpu_yaml = {
+            'custom_node_labels': {
+            'snap.com/spine': 'unknown',
+            'snap.com/region': 'us-west-2c', 
+        },
+        }
+    data.update(gpu_yaml) 
+    return data
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description="Script to generate yaml file for AWS Pytorch Job")
+    parser.add_argument('--yaml_folder', type=str, default='./', help='path to save the yaml folder')
+    parser.add_argument('--gpu_memory', type=int, default=40, help='GPU memory (in GB)')
+    parser.add_argument('--gpu_num', type=int, default=1, help='Number of GPUs')
+    parser.add_argument('--cpu_num_per_gpu', type=int, default=6, help='Number of CPUs per GPU')
+    parser.add_argument('--memory_per_gpu', type=float, default=30.0, help='Memory per GPU')
+    parser.add_argument('--replicas', type=int, default=1, help='Number of replicas')
+    parser.add_argument('--project_name', type=str, default='magic123', help='Project name')
+    parser.add_argument('--project_support_alias', type=str, default='img2mesh', help='Project support alias')
+    #parser.add_argument('--pre_run_event', type=str, default='export PATH=/nfs/code/gqian/miniconda3/bin:$PATH && conda init bash && source ~/.bashrc && cd /nfs/code/gqian/img2mesh ', help='Pre-run event command')
+    parser.add_argument('--pre_run_event', type=str, default='cd /nfs/code/gqian/img2mesh', help='Pre-run event command')
+    parser.add_argument('--command', type=str, default='while :; do sleep 1000; done', help='Command')
+    parser.add_argument('--job_name', type=str, default='debug', help='Job name')
+    parser.add_argument('--force_node', action='store_true',
+                         help="use normal cluster not debug cluster")
+    args, unknown = parser.parse_known_args()
+    args.job_name = args.job_name.replace('_', '-').replace('.', '-')[:51] # do not support _ and . in job name, and max length is limited (around 70)
+
+    # "bash scripts/magic123/run_single_bothpriors.sh 0 r256 data/nerf4/drums rgba.png --h 300 --w 300"
+    data = generate_yaml(**vars(args))
+    yaml_str = yaml.safe_dump(data)
+
+    # Write the YAML content to a file
+    os.makedirs(args.yaml_folder, exist_ok=True)
+    yaml_path = os.path.join(args.yaml_folder, f'{args.job_name}.yaml')
+    with open(yaml_path, 'w') as file:
+        file.write(yaml_str)
+    print(f'YAML file saved to {yaml_path}')
+    
+    # launch the job using snap_rutls
+    os.system(f'yes yes | snap_rutils cluster run {yaml_path} -s')
+    
+    # show the job status
+    os.system(f'kubectl get pods | grep {args.job_name} ')
+
+    # show the job logs
+    os.system(f'kubectl logs {args.job_name}-worker-0')

scripts/snap/submit_cluster_job.sh

@@ -0,0 +1,18 @@
+script_name=$1
+runid=$2
+runid2=$3
+i=$4
+imagename=$5
+run1=$6
+run2=$7
+arguments="${@:8}"
+
+timestamp=$(date +'%Y%m%d')
+[ -d "$i" ] && echo "$i exists."
+example=$(basename $i)
+echo ${@:8}
+python scripts/snap/submit_cluster_job.py --yaml_folder scripts/snap/yamls \
+        --gpu_memory 40 --gpu_num 1 --force_node --cpu_num_per_gpu 6 --memory_per_gpu 30.0 --replicas 1 \
+        --project_name magic123 --project_support_alias img2mesh \
+        --job_name gqian-$timestamp-$runid-$runid2-$example \
+        --command "bash $script_name 0 $runid $runid2 $i $imagename $run1 $run2 $arguments "

scripts/texural_inversion/textural_inversion.sh

@@ -0,0 +1,52 @@
+#! /bin/bash
+#SBATCH -N 1
+#SBATCH --array=0
+#SBATCH -J dreamfusion
+#SBATCH -o slurm_logs/%x.%3a.%A.out
+#SBATCH -e slurm_logs/%x.%3a.%A.err
+#SBATCH --time=9:00:00
+#SBATCH --gres=gpu:v100:1
+#SBATCH --cpus-per-gpu=6
+#SBATCH --mem=30G
+##SBATCH --gpus=1
+
+module load gcc/7.5.0
+
+echo "===> Anaconda env loaded"
+#source ~/.bashrc
+#source activate magic123
+source venv_magic123/bin/activate 
+
+nvidia-smi
+nvcc --version
+
+hostname
+NUM_GPU_AVAILABLE=`nvidia-smi --query-gpu=name --format=csv,noheader | wc -l`
+echo "number of gpus:" $NUM_GPU_AVAILABLE
+
+
+MODEL_NAME=$2 # "path-to-pretrained-model" runwayml/stable-diffusion-v1-5
+DATA_DIR=$3 # "path-to-dir-containing-your-image"
+OUTPUT_DIR=$4 # "path-to-desired-output-dir"
+placeholder_token=$5 # _ironman_
+init_token=$6 # ironman
+
+# run texturaal inversion
+CUDA_VISIBLE_DEVICES=$1 python textual-inversion/textual_inversion.py \
+  --pretrained_model_name_or_path=$MODEL_NAME \
+  --train_data_dir=$DATA_DIR \
+  --learnable_property="object" \
+  --placeholder_token=$placeholder_token \
+  --initializer_token=$init_token \
+  --resolution=512 \
+  --train_batch_size=16 \
+  --gradient_accumulation_steps=1 \
+  --max_train_steps=3000 \
+  --lr_scheduler="constant" \
+  --lr_warmup_steps=0 \
+  --output_dir=$OUTPUT_DIR \
+  --use_augmentations \
+  ${@:7}
+
+# test textural inversion
+CUDA_VISIBLE_DEVICES=$1 python guidance/sd_utils.py --text "A high-resolution DSLR image of <token>" --learned_embeds_path $OUTPUT_DIR  --workspace $OUTPUT_DIR 

scripts/texural_inversion/textural_inversion_list.sh

@@ -0,0 +1,13 @@
+
+
+examples=(
+	'data/nerf4/chair'
+	'data/nerf4/drums'
+	'data/nerf4/ficus'
+	'data/nerf4/mic'
+)
+
+for i in "${examples[@]}"; do
+	filename=$(basename "$i")
+	bash scripts/texural_inversion/textural_inversion.sh 0 runwayml/stable-diffusion-v1-5 "$i"/rgba.png out/texural_inversion/${filename} _nerf_${filename}_ ${filename} --max_train_steps 3000
+done