Roman Solovyev

I used the following format of files for training: Files with annotations: validation-annotations-bbox-level-1.csv and train-annotations-bbox-level-1.csv ``` ImageID,Source,LabelName,Confidence,XMin,XMax,YMin,YMax,IsOccluded,IsTruncated,IsGroupOf,IsDepiction,IsInside 0001eeaf4aed83f9,freeform,/m/0cmf2,1.0,0.022463999999999998,0.9641780000000001,0.070656,0.800164,0,0,0,0,0 00075905539074f2,freeform,/m/0dzf4,1.0,0.015193,0.11075299999999999,0.251654,0.367367,0,0,0,1,0 00075905539074f2,freeform,/m/0dzf4,1.0,0.159303,0.257175,0.116079,0.21883899999999998,0,0,0,1,0 00075905539074f2,freeform,/m/0dzf4,1.0,0.33115700000000003,0.46601899999999996,0.070312,0.17739000000000002,0,0,0,1,0 00075905539074f2,freeform,/m/0dzf4,1.0,0.518423,0.644808,0.061676999999999996,0.219703,0,0,0,1,0 00075905539074f2,freeform,/m/0dzf4,1.0,0.721102,0.905286,0.187752,0.43126800000000004,0,0,0,1,0 00075905539074f2,freeform,/m/0dzf4,1.0,0.8844780000000001,0.969249,0.558207,0.797405,0,0,0,1,0 0007cebe1b2ba653,freeform,/m/035r7c,1.0,0.732595,0.8215790000000001,0.026793,0.403305,0,0,0,0,0 0007cebe1b2ba653,freeform,/m/035r7c,1.0,0.826202,0.931365,0.0068390000000000005,0.375544,0,0,0,0,0 0007cebe1b2ba653,freeform,/m/0bt9lr,1.0,0.420795,0.79402,0.181372,0.7205739999999999,0,0,0,0,0 0007cebe1b2ba653,freeform,/m/0dzf4,1.0,0.777104,0.804149,0.000132,0.01991,0,1,0,0,0 0007d6cf88afaa4a,freeform,/m/0bt9lr,1.0,0.173566,0.9025690000000001,0.216627,0.941628,0,0,0,0,0 0008e425fb49a2bf,freeform,/m/0bt9lr,1.0,0.22699699999999998,0.715052,0.11206400000000001,0.934448,0,0,0,0,0...

Can you post exact time?

1) You need to ensure GPU is used 2) Long time is possible for first recognition - because of long model initialization. Try to recognize several images and check how...

Did you load model before each image? Or use the same?

That's strange. It shouldn't take more than a second. Which tensorflow and keras version you use? Did you try resnet101 and resnet50?

Looks fine. Do you use model for inference? Try model based on resnet50 and check timing.

Sorry don't know what it can be ( Did you try script with inference example? https://github.com/ZFTurbo/Keras-RetinaNet-for-Open-Images-Challenge-2018/blob/master/retinanet_inference_example.py

During this 60 seconds for inference can you check GPU usage?

I don't think it's possible to reduce load model time. But you can load model once and keep it in memory while processing images.

Could you please provide additional details, because if you use projects with totally same connections you won't have any problem with Pin Planner. Do you have some kind of error...