Upload run_varsr_offline_3dsr.sh

run_varsr_offline_3dsr.sh

@@ -0,0 +1,84 @@
+#!/bin/bash
+set -e
+
+######################################################################
+# User-configurable parameters
+######################################################################
+dataset_name="mipnerf360" # choose from [mipnerf360, llff]
+dataset_path="/home/shulei/3D-SR-AR/dataset/mipnerf"
+varsr_root="/home/shulei/3D-SR-AR/AR/VARSR/outputs/mipnerf360/load_DS_16"
+gpu=0
+HR_factor=4
+GS_iters=5000
+DENSIFY_END=2500
+
+# Pretrained LR model path. This should be the same LR checkpoint used by
+# the StableSR baseline for a fair comparison.
+output_dir="/home/shulei/3DSR-outputs/outputs/LR_pretrained/input_DS_$((HR_factor * 4))"
+
+# VARSR-driven experiment directory.
+exp_dir="/home/shulei/3D-SR-AR/AR/VARSR/outputs/${dataset_name}/varsr_load_DS_$((HR_factor * 4))"
+######################################################################
+
+if [ "$dataset_name" == "llff" ]; then
+  scenes=(
+    "fern"
+    "flower"
+    "fortress"
+    "horns"
+    "leaves"
+    "orchids"
+    "room"
+    "trex"
+  )
+else
+  scenes=(
+    "bicycle"
+  #  "flowers"
+   # "garden"
+  #  "stump"
+   # "treehill"
+    #"room"
+   # "counter"
+   # "kitchen"
+   # "bonsai"
+  )
+fi
+
+for scene in "${scenes[@]}"; do
+    echo "Starting VARSR-offline 3DSR refinement: ${scene}"
+    export NUM_ITERS=${GS_iters}
+
+    OMP_NUM_THREADS=4 CUDA_VISIBLE_DEVICES=$gpu python train_3dsr.py \
+        -s ${dataset_path}/${scene} \
+        -m ${output_dir}/${scene} \
+        -r $HR_factor \
+        --port $((6000 + gpu)) \
+        --eval \
+        --output_folder ${exp_dir} \
+        --load_pretrain \
+        --sr_backend offline \
+        --offline_sr_dir ${varsr_root}/${scene} \
+        --outdir ${exp_dir}/${scene} \
+        --seed 42 \
+        --fidelity_train_en \
+        --wt_lr 1 \
+        --kernel_size 0.1 \
+        --skip_train_results \
+        --densify_end ${DENSIFY_END}
+
+    echo "Rendering scene: ${scene}"
+    OMP_NUM_THREADS=4 CUDA_VISIBLE_DEVICES=${gpu} python render.py \
+        -m ${exp_dir}/${scene} \
+        -r $HR_factor \
+        --data_device cpu \
+        --skip_train
+
+    echo "Evaluating scene: ${scene}"
+    OMP_NUM_THREADS=4 CUDA_VISIBLE_DEVICES=${gpu} python metrics.py \
+        -m ${exp_dir}/${scene} \
+        -g ${dataset_path}/${scene}
+
+    echo "Completed VARSR-offline scene: ${scene}"
+    echo "------------------------------------------------------"
+done