Remove the FPN from the RetinaNet with MobileNetV3 backbone

[datumbox]

RetinaNet + MobileNetV3 large

Trained using the code committed at 110b016.

The current temporary pre-trained model was trained:

python -m torch.distributed.launch --nproc_per_node=8 --use_env train.py\
 --dataset coco --model retinanet_mobilenet_v3_large --epochs 26 --lr-steps 16 22\
 --aspect-ratio-group-factor 3 --lr 0.01

Submitted batch job 34681346

Epoch 22
0: Test: Total time: 0:00:24 (0.0395 s / it)
0: Averaged stats: model_time: 0.0171 (0.0188)  evaluator_time: 0.0101 (0.0166)
0: Accumulating evaluation results...
0: DONE (t=10.58s).
0: IoU metric: bbox
0:  Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.221
0:  Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.387
0:  Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.220
0:  Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.053
0:  Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.231
0:  Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.373
0:  Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.219
0:  Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.331
0:  Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.352
0:  Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.121
0:  Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.389
0:  Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.563
0: Training time 4:59:35

Speed benchmark:
0.62 sec per image on CPU