)]}'
{
  "commit": "89e832b90619510b0f107d0946af5c1374fc0f1b",
  "tree": "338731db5854a07ff7687b41d5e190bb646b7b36",
  "parents": [
    "9284af62c37d62117c007d100e0442f144220ab8"
  ],
  "author": {
    "name": "Urvang Joshi",
    "email": "urvang@google.com",
    "time": "Tue Nov 20 15:28:54 2018 -0800"
  },
  "committer": {
    "name": "Urvang Joshi",
    "email": "urvang@google.com",
    "time": "Thu Nov 22 00:38:47 2018 +0000"
  },
  "message": "Rate-control bugfix for golden frame in CQ mode.\n\nEarlier, here is what was happening for the golden frame in CQ mode:\n- The active_best_quality was set to \u0027rc-\u003earf_q\u0027\n- But in some cases, \u0027rc-\u003earf_q\u0027 was never really set (e.g. when\ncpi-\u003enew_bwdref_update_rule \u003d\u003d 0), and so it still had the initial value 0.\n- This meant that for easy clips / high target bitrates, golden frame\nended up choosing q \u003d 0 (or some value close to 0), and so it was really\nlarge.\n\nThis behavior was simply a bug. \u0027active_best_quality\u0027 should be set from\n\u0027rc-\u003earf_q\u0027 only for internal ARFs with cpi-\u003enew_bwdref_update_rule \u003d\u003d 1\n(as we would have set the \u0027rc-\u003earf_q\u0027 value already by then, when we\nencoded the ARF frame before this internal ARF).\n\nThis patch corrects this behavior, and also adds a relevant assertion to\nensure that the behavior is tested.\n\nIn an extreme case, for example, here is how golden frame compares:\n\nBefore: 328776 bytes, PSNR \u003d 100 (lossless)\nAfter: 680 bytes, PSNR \u003d 56.753\n\nSTATS_CHANGED for CQ mode\n\nChange-Id: I47697768dda1196bd0a95eec105179b332dd9222\n",
  "tree_diff": [
    {
      "type": "modify",
      "old_id": "56203cd7aa4d1d0f7e6edadc0fc616a4aaa2e4d0",
      "old_mode": 33188,
      "old_path": "av1/encoder/ratectrl.c",
      "new_id": "21632c0ef93e18f58a011b7e0e00c6a54975257e",
      "new_mode": 33188,
      "new_path": "av1/encoder/ratectrl.c"
    }
  ]
}