RC运行记录--仅测试

基于学习的图像压缩模型运行记录

RC: Learning Better Lossless Compression Using Lossy Compression

---

---

前言

文章：Learning Better Lossless Compression Using Lossy Compression
代码：github
运行RC需要安装bpg。安装步骤

---

一、下载代码

下载代码前需要创建目录如下：

export RC_ROOT="/path/to/wherever/you/want"  #export RC_ROOT="/mnt/disk1/lsj/RC"
mkdir -p "$RC_ROOT"
mkdir -p "$RC_ROOT/datasets"
mkdir -p "$RC_ROOT/models"
pushd $RC_ROOT
git clone https://github.com/fab-jul/RC-PyTorch

二、测试BPG

pushd $RC_ROOT/RC-PyTorch/src
bash test_bpg_available.sh

正常返回：

三、创建环境

NAME=pyt11  # You can change this to whatever you want.
conda create -n $NAME python==3.7 pip -y
conda activate $NAME

# Note that we're using PyTorch 1.1 and CUDA 10.0. 
# Other combinations may also work but have not been tested!
pip install torch==1.11.0+cu113 torchvision==0.12.0+cu113 -f https://download.pytorch.org/whl/torch_stable.html

# Install the pip requirements
pushd $RC_ROOT/RC-PyTorch/src
pip install -r requirements.txt

四、测试压缩图片

1.下载模型文件

MODELS_DIR="$RC_ROOT/models"
bash get_models.sh "$MODELS_DIR"

2.数据预处理（BPG）

数据放在"$RC_ROOT/datasets"下

pushd "$RC_ROOT/RC-PyTorch/src"
bash prep_bpg_ds.sh A9_17 $RC_ROOT/datasets/Kodak  #我这里用Kodak作为测试数据集

3.测试bpsp

DATASET_DIR="$RC_ROOT/datasets"
MODELS_DIR="$RC_ROOT/models"

pushd "$RC_ROOT/RC-PyTorch/src"

# Note: depending on your environment, adapt CUDA_VISIBLE_DEVICES.
CUDA_VISIBLE_DEVICES=0 python -u run_test.py \
    "$MODELS_DIR" 1109_1715 "AUTOEXPAND:$DATASET_DIR/Kodak" \
    --restore_itr 1000000 \
    --tau \
    --clf_p "$MODELS_DIR/1115_1729*/ckpts/*.pt" \
    --qstrategy CLF_ONLY

clf_p要改成模型文件路径
运行报错修改代码：run_test.py。在parse_flags(args):函数里的if flags.clf_p语句下插入

print(f'Using Q classifier {flags.clf_p}')
matched_files = glob.glob(flags.clf_p)
assert matched_files, f"No files found for pattern: {flags.clf_p}"
if len(matched_files) > 1:
        print(f"Multiple files found for pattern: {flags.clf_p}")
flags.clf_p = matched_files[0]

结果

---