Tensorflow Research Cloud granted me 100+5 TPUs for 30 days. I am using TPUs to train deep neural networks for Object Detection task.
| Model name | Speed (ms, TITAN X) | COCO mAP | FLIR ADAS | Open Images V4 | ERL2018 |
|---|---|---|---|---|---|
| ssd_mobilenet_v1_0.75_depth_coco | 26 | 18 | |||
| ssd_mobilenet_v1_quantized_coco | 29 | 18 | |||
| ssd_mobilenet_v1_0.75_depth_quantized_coco | 29 | 16 | |||
| ssd_mobilenet_v1_ppn_coco | 26 | 20 | |||
| ssd_mobilenet_v1_fpn_coco | 56 | 32 | |||
| ssd_resnet_50_fpn_coco | 76 | 35 |
With the script download_tpu_models_to_gs.sh you can download all models from the table above and upload them into your storage bucket.