Added PILToTensor and ConvertImageDtype classes in reference scripts

[prabhat00155]

Detection:

srun -p train --cpus-per-task=16 -t 10:00:00 --gpus-per-node=8 python -m torch.distributed.launch --nproc_per_node=8 --use_env train.py --model fasterrcnn_resnet50_fpn --output-dir logs/to_tensor/ --pretrained --test-only

Without these changes:

IoU metric: bbox
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.369
 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.585
 Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.396
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.212
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.403
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.482
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.307
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.485
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.509
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.317
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.544
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.649

With these changes:

IoU metric: bbox
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.369
 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.585
 Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.396
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.212
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.403
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.482
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.307
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.485
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.509
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.317
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.544
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.649

Segmentation:

srun -p train --cpus-per-task=16 -t 10:00:00 --gpus-per-node=8 python -m torch.distributed.launch --nproc_per_node=8 --use_env train.py --model fcn_resnet50 --output-dir logs/to_tensor/ --pretrained --test-only

Without these changes:

global correct: 91.4
average row correct: ['94.0', '80.9', '73.9', '72.3', '49.9', '56.9', '81.0', '66.5', '88.6', '43.2', '75.7', '67.6', '80.9', '84.3', '81.2', '90.9', '49.0', '80.4', '61.2', '77.7', '69.0']
IoU: ['90.3', '72.6', '61.2', '61.6', '42.7', '46.9', '73.0', '53.7', '80.3', '33.6', '64.1', '33.9', '61.2', '72.0', '73.6', '80.7', '29.1', '63.5', '48.2', '71.2', '56.3']
mean IoU: 60.5

With these changes:

Test: Total time: 0:00:49
global correct: 91.4
average row correct: ['94.0', '80.9', '73.9', '72.3', '49.9', '56.9', '81.0', '66.5', '88.6', '43.2', '75.7', '67.6', '80.9', '84.3', '81.2', '90.9', '49.0', '80.4', '61.2', '77.7', '69.0']
IoU: ['90.3', '72.6', '61.2', '61.6', '42.7', '46.9', '73.0', '53.7', '80.3', '33.6', '64.1', '33.9', '61.2', '72.0', '73.6', '80.7', '29.1', '63.5', '48.2', '71.2', '56.3']
mean IoU: 60.5

cc @vfdev-5 @datumbox

[datumbox]

Thanks @prabhat00155, the numbers look OK though it would be good to provide also runs on the latest main to confirm.

For example on our docs we report that the mAP of the Faster R-CNN ResNet-50 FPN model should e 37.0 not 36.9 that you got but I believe this is the result of a separate accuracy deterioration and not related to your changes. By having runs on the latest main you can prove that this PR is not to blame.

Finally below I left some comments for your consideration, let me know what you think:

[datumbox]

LGTM, thanks!