)]}'
{
  "commit": "d9b9ebabace076d485f7d7b7d98fe6ff0a481f64",
  "tree": "914437bbaf34b680962f9c79d792f58b511b43f8",
  "parents": [
    "23ebc8777b81a01a64cf669cdb25448004f0bc9a"
  ],
  "author": {
    "name": "Hui Su",
    "email": "huisu@google.com",
    "time": "Thu May 03 12:09:45 2018 -0700"
  },
  "committer": {
    "name": "Hui Su",
    "email": "huisu@google.com",
    "time": "Tue May 15 21:47:41 2018 +0000"
  },
  "message": "Rework the ML based transform split pruning feature\n\n-Use mean and standard deviation of the residue block and sub-blocks as\nfeature.\n-Train neural nets model for different transform block sizes.\n-Horizontal and vertical rectangular blocks share the same model. For\nexample, TX_4X8 and TX_8X4 use the same model; TX_4X16 and TX_16X4 use\nthe same model.\n\nCompression loss(30 frames) is under 0.05% for both speed 0 and 1.\n\nEncoding speedup varies among different sequences; no speed regression\nwas observed. Some samples:\n\nSpeed 0(15 frames):\nbasketballpass_240p     8%\nbowing_cif             10%\nbqsquare_240p          15%\ncheer_sif              22%\n\nSpeed 1(30 frames):\nblowingbubbles_240p   4.5%\ncheer_sif              10%\nflower_cif              4%\nfootball_cif            7%\n\nChange-Id: If102fa333d30fe7e5f5b1b302e92bf688fc0bbf8\n",
  "tree_diff": [
    {
      "type": "modify",
      "old_id": "faa8aafdede67e2b7232407fecf3170e80f28f04",
      "old_mode": 33188,
      "old_path": "av1/encoder/rdopt.c",
      "new_id": "dc2a3a586d1edaea178cdcc10ae3a0c6cda35503",
      "new_mode": 33188,
      "new_path": "av1/encoder/rdopt.c"
    },
    {
      "type": "modify",
      "old_id": "ce6d72f9b46cfe3d5def1536affcab7b02f5bc85",
      "old_mode": 33188,
      "old_path": "av1/encoder/speed_features.c",
      "new_id": "46beaeb912c7598bae59d9ff53fdd8cb21623f35",
      "new_mode": 33188,
      "new_path": "av1/encoder/speed_features.c"
    },
    {
      "type": "modify",
      "old_id": "f88bcfd265e5c823247da83fb0ac75b15e783aee",
      "old_mode": 33188,
      "old_path": "av1/encoder/speed_features.h",
      "new_id": "b0921ab00dbb65e58b0743cb678f886ce25881ed",
      "new_mode": 33188,
      "new_path": "av1/encoder/speed_features.h"
    },
    {
      "type": "modify",
      "old_id": "119456297612a31d4b8d268c0f316ee8173e2a70",
      "old_mode": 33188,
      "old_path": "av1/encoder/tx_prune_model_weights.h",
      "new_id": "69063b80154b0dce82d6275a490bb93a3134ae55",
      "new_mode": 33188,
      "new_path": "av1/encoder/tx_prune_model_weights.h"
    }
  ]
}