Tensorflow Custom Object Detection Training Speed

Question

I tried to train a frcnn for detecting money it has 3 classes.

I used this link https://github.com/EdjeElectronics/TensorFlow-Object-Detection-API-Tutorial-Train-Multiple-Objects-Windows-10 to generate something. And it uses faster_rcnn_inception_v2_coco_2018_01_28 config

It takes too much time how can i fix this or is this normal ?

INFO:tensorflow:global_step/sec: 0.049895
INFO:tensorflow:Recording summary at step 1830.
INFO:tensorflow:Recording summary at step 1830.
INFO:tensorflow:global step 1831: loss = 0.2820 (22.984 sec/step)
INFO:tensorflow:global step 1831: loss = 0.2820 (22.984 sec/step)
INFO:tensorflow:global step 1832: loss = 0.2418 (``20.894 sec/step)

and keep going like that times

my tensorflow version b'v1.9.0-0-g25c197e023' 1.9.0 Cuda compilation tools, release 9.1, V9.1.85

My computer has 8gb ram GTX 950M and 4gb Memory 2.6Ghz i7-6700HQ

Answer 1

Ok got it.

CUDA 9.1 is not compatible Tensorflow GPU so I didn't use Tf-GPU