tmp.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "def load_binary_masks(bin_file_path):\n",
    "    with open(bin_file_path, 'rb') as f:\n",
    "        shape = np.fromfile(f, dtype=np.int32, count=3)\n",
    "        # 读取掩码数据，使用uint16类型\n",
    "        binary_masks = np.fromfile(f, dtype=np.uint16)\n",
    "        binary_masks = binary_masks.reshape(shape)\n",
    "    return binary_masks"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "orig= load_binary_masks('/nfs/dataset-ofs-voyager-research/shichen/project/video_diffusion/ConsisID/workdirs/step4_track_masks_data/515d576284baf2cb5ecc534f3105f3fb_0_107/tracking_mask_results/1/masks.bin')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "orig = orig.astype(np.uint8)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import blosc\n",
    "import os\n",
    "\n",
    "def compress_3d_direct(binary_volume, output_file='3d_direct.bin'):\n",
    "    \"\"\"直接压缩整个3D体积\"\"\"\n",
    "    # 先用np.packbits进行基础压缩\n",
    "    packed = np.packbits(binary_volume)\n",
    "    \n",
    "    # 使用专门的压缩算法\n",
    "    compressed = blosc.compress(packed.tobytes(), \n",
    "                               typesize=1, \n",
    "                               cname='zstd', \n",
    "                               clevel=9,\n",
    "                               shuffle=blosc.BITSHUFFLE)  # 位级混排，对二值数据更有效\n",
    "    \n",
    "    # 保存压缩数据和元数据\n",
    "    with open(output_file, 'wb') as f:\n",
    "        # 保存元数据（形状）\n",
    "        shape_info = np.array(binary_volume.shape, dtype=np.int32)\n",
    "        f.write(shape_info.tobytes())\n",
    "        # 保存压缩数据\n",
    "        f.write(compressed)\n",
    "    \n",
    "    # 计算压缩比\n",
    "    original_size = binary_volume.nbytes\n",
    "    compressed_size = os.path.getsize(output_file)\n",
    "    return original_size, compressed_size, original_size/compressed_size"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(99688320, 121135, 822.9522433648409)"
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "compress_3d_direct(orig)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import blosc\n",
    "import os\n",
    "\n",
    "def decompress_3d_direct(input_file='3d_direct.bin'):\n",
    "    \"\"\"解压缩由compress_3d_direct压缩的3D二值数组\"\"\"\n",
    "    with open(input_file, 'rb') as f:\n",
    "        # 读取元数据（形状信息）\n",
    "        # 假设形状是3维的，每个维度是32位整数(4字节)\n",
    "        shape_bytes = f.read(3 * 4)  # 3个int32\n",
    "        shape_info = np.frombuffer(shape_bytes, dtype=np.int32)\n",
    "        \n",
    "        # 读取压缩数据\n",
    "        compressed_data = f.read()\n",
    "        \n",
    "        # 解压缩blosc数据\n",
    "        decompressed_bytes = blosc.decompress(compressed_data)\n",
    "        \n",
    "        # 将字节转换回numpy数组（仍是打包的位）\n",
    "        packed_array = np.frombuffer(decompressed_bytes, dtype=np.uint8)\n",
    "        \n",
    "        # 计算原始数组中的元素总数\n",
    "        total_elements = shape_info[0] * shape_info[1] * shape_info[2]\n",
    "        \n",
    "        # 解开位打包，还原为布尔数组\n",
    "        unpacked = np.unpackbits(packed_array)\n",
    "        \n",
    "        # 可能需要截断多余的位（unpackbits总是产生8的倍数长度）\n",
    "        if len(unpacked) > total_elements:\n",
    "            unpacked = unpacked[:total_elements]\n",
    "        \n",
    "        # 重塑为原始形状\n",
    "        result = unpacked.reshape(tuple(shape_info)).astype(np.bool_)\n",
    "        \n",
    "        return result\n",
    "\n",
    "# 验证压缩和解压是否正确\n",
    "def verify_compression(original_array, input_file='3d_direct.bin'):\n",
    "    \"\"\"验证压缩和解压是否无损\"\"\"\n",
    "    # 解压缩\n",
    "    decompressed = decompress_3d_direct(input_file)\n",
    "    \n",
    "    # 检查形状是否相同\n",
    "    shape_match = original_array.shape == decompressed.shape\n",
    "    \n",
    "    # 检查内容是否相同\n",
    "    content_match = np.array_equal(original_array, decompressed)\n",
    "    \n",
    "    print(f\"形状匹配: {shape_match}\")\n",
    "    print(f\"内容匹配: {content_match}\")\n",
    "    \n",
    "    if not content_match:\n",
    "        # 找出不匹配的元素数量\n",
    "        diff_count = np.sum(original_array != decompressed)\n",
    "        total_elements = np.prod(original_array.shape)\n",
    "        print(f\"不匹配元素: {diff_count}/{total_elements} ({diff_count/total_elements*100:.6f}%)\")\n",
    "    \n",
    "    return shape_match and content_match"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "tmp = decompress_3d_direct()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 32,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "(tmp.astype(np.uint8) == orig).all()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "dino",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.19"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}