| sample_dir="evals-release/clipb_test" | |
| export OMP_NUM_THREAD | |
| torchrun \ | |
| --nproc_per_node=... \ | |
| --master_addr=... --master_port=... \ | |
| tokenizer/reconstruction_vq_ddp.py \ | |
| --sample-dir=$sample_dir \ | |
| --per-proc-batch-size 32 \ | |
| --codebook-size=32768 \ | |
| --image-size 256 \ | |
| --image-size-eval 256 \ | |
| --infer_interpolate \ | |
| --enhanced_decoder \ | |
| --teacher "clipb_224" \ | |
| --vq-ckpt="/path/to/tokenflow_clipb_32k_enhanced.pt" && | |
| src_dir1=$sample_dir/samples/ && | |
| src_dir2=$sample_dir/gts/ && | |
| python3 -m evaluations.vq.pytorch_fid $src_dir1 $src_dir2 --device cuda:0 --results_path $sample_dir | |
| set +x | |
| target_folder_path=$sample_dir/demos/ | |
| mkdir -p "$target_folder_path" | |
| files=() | |
| for file in $(ls "$src_dir1"); do | |
| if [[ "$file" == *.png ]]; then | |
| files+=("$file") | |
| fi | |
| done | |
| for ((i=0; i<5 && i<${#files[@]}; i++)); do | |
| selected_file="${files[$i]}" | |
| cp "$src_dir1/$selected_file" "$target_folder_path" | |
| done | |
| total_files=${#files[@]} | |
| for ((i=total_files-5; i<total_files && i>=0; i++)); do | |
| selected_file="${files[$i]}" | |
| cp "$src_dir1/$selected_file" "$target_folder_path" | |
| done | |
| echo "Selected files have been copied to $target_folder_path" |