)]}'
{
  "commit": "2d1ead342cbaaba5f5fd07c3165f48e1029d327c",
  "tree": "243abfb4d66c22922d3bf95055cb61bb0b249e70",
  "parents": [
    "21108d800c605a1e63cfa9f27a333e7b46924d84"
  ],
  "author": {
    "name": "Yaowu Xu",
    "email": "yaowu@google.com",
    "time": "Mon Feb 06 15:10:13 2012 -0800"
  },
  "committer": {
    "name": "Paul Wilkins",
    "email": "paulwilkins@google.com",
    "time": "Mon Feb 13 13:20:02 2012 +0000"
  },
  "message": "Changed how coefficient probability table is updated\n\nAdded a frame level flag to indicate if coef probabilities are updated\nat all for the frame.\n\nDuring the experimental work with 8x8 transform, it is discovered that\neven in the case of no probability is ever update, cost of transmitting\n\"no update\" for each of probabilities can run up to become a significant\noverhead cost. A single bit to indicate no-update for all coef probs\nis therefore helpful, which is also demonstrated by the test results:\n\n1. On Cif set:\nhttp://www.corp.google.com/~yaowu/no_crawl/t8x8/cif_t8x8_updprob.html\n(avg psnr: .14%, glb psnr: .14% SSIM: .13%)\n\n2. On HD set:\nhttp://www.corp.google.com/~yaowu/no_crawl/t8x8/HD_t8x8_updprob.html\n(avg psnr: .02%  glb psnr: .01% SSIM: .02%)\nIt should be noted that the gain on HD is smaller because the average bit\nrate is much higher in contrast to the overhead bit cost.\n\nChange-Id: I46db270e693ee8799fef34a14d8260868ce4cd16\n",
  "tree_diff": [
    {
      "type": "modify",
      "old_id": "64b58b5bb7c94f943f7b8d1856a8d5b7844b4c4f",
      "old_mode": 33188,
      "old_path": "vp8/decoder/decodframe.c",
      "new_id": "c5385a022c02d1b667eefeeff23102b6eca64dd7",
      "new_mode": 33188,
      "new_path": "vp8/decoder/decodframe.c"
    },
    {
      "type": "modify",
      "old_id": "f3f0c1ba2abc33235d08ee478f069479ee59ac83",
      "old_mode": 33188,
      "old_path": "vp8/encoder/bitstream.c",
      "new_id": "2eb959d795bd8e694e3199154f0886a047bc6f3b",
      "new_mode": 33188,
      "new_path": "vp8/encoder/bitstream.c"
    }
  ]
}