Speedup expand when meets large image

[Sundrops]

ref: numpy/numpy#13001

%%timeit
a = np.full((4000, 2000, 3), [0.0, 0.0, 0.0], np.float32)
76.9 ms ± 346 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)
%%timeit
a = np.full((4000, 2000, 3), [0, 0, 0], np.float32)
71.4 ms ± 516 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)
%%timeit
a = np.full((4000, 2000, 3), 0.0, np.float32)
47.2 ms ± 1.21 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
%%timeit
a = np.full((4000, 2000, 3), 0, np.float32)
47 ms ± 139 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)
%%timeit
a = np.empty((4000, 2000, 3), np.float32)
a.fill(0.0)
46.3 ms ± 278 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)
%%timeit
a = np.empty((4000, 2000, 3), np.float32)
a.fill(0)
46.3 ms ± 184 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)

[codecov]

Codecov Report

Merging #4089 (1b35880) into master (7ad44cd) will increase coverage by 0.02%.
The diff coverage is 20.00%.

@@            Coverage Diff             @@
##           master    #4089      +/-   ##
==========================================
+ Coverage   61.92%   61.95%   +0.02%     
==========================================
  Files         225      225              
  Lines       16641    16647       +6     
  Branches     2814     2817       +3     
==========================================
+ Hits        10305    10313       +8     
+ Misses       5838     5832       -6     
- Partials      498      502       +4     

Flag	Coverage Δ
unittests	61.90% <20.00%> (+0.01%)	⬆️

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

Impacted Files	Coverage Δ
mmdet/core/evaluation/mean_ap.py	41.63% <16.66%> (-0.55%)	⬇️
mmdet/datasets/pipelines/transforms.py	73.18% <25.00%> (-0.31%)	⬇️
mmdet/core/bbox/assigners/max_iou_assigner.py	72.36% <0.00%> (-1.32%)	⬇️
mmdet/models/dense_heads/corner_head.py	76.77% <0.00%> (+1.91%)	⬆️
mmdet/models/detectors/cornernet.py	100.00% <0.00%> (+5.12%)	⬆️

---

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 7ad44cd...6a21820. Read the comment docs.

[hellock]

• We can leave a comment here to tell users why.
• The processed image does not necessarily have 3 channels, we can use if np.all(self.mean == self.mean[0]) instead.

[Sundrops]

Done