Redundant mesh grid creation for VSI

[denproc]

Creation of a mesh grid for a high-volume datasets with high-res images could slow down the performance computing VSI.

In case of VSI, one mesh grid is created in sdsp, while another one is generated in _log_gabor. As it was shown in #266 and #334, resolution of a similar issue improved the performance of the FSIM computation.

Therefore, I suggest two ideas to tackle the issue:

1. Local: reuse a mesh grid from sdsp for filter generation.
2. Global: mesh grids (FSIM, VSI) could be cached similar to the strategy suggested at #227.

[denproc]

The local approach with caching the mesh grid was tested. While it halves the number of mesh grid generations, the performance improvement is marginal.

The upgraded version is available #342, but I'm not sure if it is worth merging it. @snk4tr @zakajd, what do you think?

Outcome

4 streams, no GPU

python tests/results_benchmark.py --dataset tid2013 --metrics VSI --path ../iqa_datasets/datasets/tid2013 --device cpu

# master
100%|█████████| 3000/3000 [08:01<00:00,  6.23it/s]
VSI: SRCC 0.895 KRCC 0.716

# fix/optimise_meshgrid
100%|█████████| 3000/3000 [05:53<00:00,  8.50it/s]
VSI: SRCC 0.895 KRCC 0.716

python tests/results_benchmark.py --dataset pipal --metrics VSI --path ../iqa_datasets/datasets/pipal --device cpu
# master
100%|███████| 23200/23200 [32:04<00:00, 12.06it/s]
VSI: SRCC 0.539 KRCC 0.375

# fix/optimise_meshgrid
100%|███████| 23200/23200 [33:45<00:00, 11.45it/s]
VSI: SRCC 0.539 KRCC 0.375

40 streams, Tesla V100

python tests/results_benchmark.py --dataset pipal --metrics VSI --path ./data/pipal --batch_size 16 --device coda
# Master CUDA
100%|█████████| 1450/1450 [00:42<00:00, 34.02it/s]
VSI: SRCC 0.539 KRCC 0.375

# Updated CUDA
100%|█████████| 1450/1450 [00:38<00:00, 37.72it/s]
VSI: SRCC 0.539 KRCC 0.375

python tests/results_benchmark.py --dataset pipal --metrics VSI --path ./data/pipal --batch_size 16 --device cpu
# Master CPU
100%|█████████| 1450/1450 [06:15<00:00,  3.87it/s]
VSI: SRCC 0.539 KRCC 0.375

# Updated CPU
100%|█████████| 1450/1450 [06:16<00:00,  3.86it/s]
VSI: SRCC 0.539 KRCC 0.375