[Enhance] Add waymo unit test

[Wuziyi616]

Fix issue#447. I add some basic unit tests for WaymoDataset. Mainly about getitem and show (because I want to check its compatibility in dataset_browse). I am not familiar with Waymo format of box and its evaluate function. Maybe someone can help me continue this PR?

[codecov]

Codecov Report

Merging #455 (2e55bb6) into master (b00d3dc) will increase coverage by 0.11%.
The diff coverage is n/a.

❗ Current head 2e55bb6 differs from pull request most recent head 07d0933. Consider uploading reports for the commit 07d0933 to get more accurate results

@@            Coverage Diff             @@
##           master     #455      +/-   ##
==========================================
+ Coverage   50.55%   50.67%   +0.11%     
==========================================
  Files         189      189              
  Lines       14453    14453              
  Branches     2366     2366              
==========================================
+ Hits         7307     7324      +17     
+ Misses       6655     6636      -19     
- Partials      491      493       +2     

Flag	Coverage Δ
unittests	50.67% <ø> (+0.11%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
mmdet3d/datasets/waymo_dataset.py	19.17% <0.00%> (+8.80%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update b00d3dc...07d0933. Read the comment docs.

[ZwwWayne]

The file sizes for unit tests are too big. We need to make the size of test data smaller. See other files for unit test.

[Wuziyi616]

The file sizes for unit tests are too big. We need to make the size of test data smaller. See other files for unit test.

Yes I have made them smaller (each <1MB).

[Wuziyi616]

I have some trouble in the Waymo protocol testing of test_evaluate() in test_waymo_dataset.py. I got expected mAP in the KITTI protocol, but wrong results in the Waymo protocol (all 0). The GT bbox I generated in gt.bin is the same as the predicted bbox after kitti_to_waymo conversion (here). I don't know why the results are wrong...

The predicted bbox after conversion:

objects {
  object {
    box {
      center_x: 69.67650333936074
      center_y: 33.340902794888315
      center_z: 0.7713991550544462
      width: 2.009999990463257
      length: 4.360000133514404
      height: 1.4600000381469727
      heading: -1.4807963232186179
    }
    type: TYPE_VEHICLE
  }
  score: 0.5
  context_name: "1071392229495085036_1844_790_1864_790"
  frame_timestamp_micros: 1507315488219118
}

The loaded gt.bin file GT bbox:

objects {
  object {
    box {
      center_x: 69.67650333936074
      center_y: 33.340902794888315
      center_z: 0.7713991550544462
      width: 2.009999990463257
      length: 4.360000133514404
      height: 1.4600000381469727
      heading: -1.4807963232186179
    }
    type: TYPE_VEHICLE
  }
  score: 1.0
  context_name: "1071392229495085036_1844_790_1864_790"
  frame_timestamp_micros: 1507315488219118
  camera_name: FRONT
}

[Tai-Wang]

I have some trouble in the Waymo protocol testing of test_evaluate() in test_waymo_dataset.py. I got expected mAP in the KITTI protocol, but wrong results in the Waymo protocol (all 0). The GT bbox I generated in gt.bin is the same as the predicted bbox after kitti_to_waymo conversion (here). I don't know why the results are wrong...

The predicted bbox after conversion:

objects {
  object {
    box {
      center_x: 69.67650333936074
      center_y: 33.340902794888315
      center_z: 0.7713991550544462
      width: 2.009999990463257
      length: 4.360000133514404
      height: 1.4600000381469727
      heading: -1.4807963232186179
    }
    type: TYPE_VEHICLE
  }
  score: 0.5
  context_name: "1071392229495085036_1844_790_1864_790"
  frame_timestamp_micros: 1507315488219118
}

The loaded gt.bin file GT bbox:

objects {
  object {
    box {
      center_x: 69.67650333936074
      center_y: 33.340902794888315
      center_z: 0.7713991550544462
      width: 2.009999990463257
      length: 4.360000133514404
      height: 1.4600000381469727
      heading: -1.4807963232186179
    }
    type: TYPE_VEHICLE
  }
  score: 1.0
  context_name: "1071392229495085036_1844_790_1864_790"
  frame_timestamp_micros: 1507315488219118
  camera_name: FRONT
}

Sometimes evaluation methods of these datasets may not deal with all the corner cases. You may try to validate it with more samples or add some noises to these predictions.

[Wuziyi616]

I finally solved the problem and this PR is ready for review now @ZwwWayne!