No positive ROIs

I have this message No positive ROIs at every iteration of training. Is that normal in training with TLT. Used FasterRCNN in training with RESNET18. My image size is 736x736 and object sizes are from 20x100 to 100x250.

93/4579 […] - ETA: 51:26 - rpn_cls: 0.4750 - rpn_regr: 0.0251 - detector_cls: 0.4936 - detector_regr: 0.0457No positive ROIs.
121/4579 […] - ETA: 46:55 - rpn_cls: 0.4345 - rpn_regr: 0.0252 - detector_cls: 0.4273 - detector_regr: 0.0465No positive ROIs.
164/4579 [>…] - ETA: 42:40 - rpn_cls: 0.3852 - rpn_regr: 0.0250 - detector_cls: 0.3605 - detector_regr: 0.0479No positive ROIs.
170/4579 [>…] - ETA: 42:14 - rpn_cls: 0.3793 - rpn_regr: 0.0249 - detector_cls: 0.3533 - detector_regr: 0.0480No positive ROIs.
202/4579 [>…] - ETA: 40:21 - rpn_cls: 0.3514 - rpn_regr: 0.0246 - detector_cls: 0.3203 - detector_regr: 0.0485No positive ROIs.
213/4579 [>…] - ETA: 39:51 - rpn_cls: 0.3429 - rpn_regr: 0.0246 - detector_cls: 0.3107 - detector_regr: 0.0487No positive ROIs.
228/4579 [>…] - ETA: 39:13 - rpn_cls: 0.3322 - rpn_regr: 0.0245 - detector_cls: 0.2989 - detector_regr: 0.0490No positive ROIs.
266/4579 [>…] - ETA: 37:53 - rpn_cls: 0.3083 - rpn_regr: 0.0244 - detector_cls: 0.2738 - detector_regr: 0.0498No positive ROIs.
366/4579 [=>…] - ETA: 35:21 - rpn_cls: 0.2620 - rpn_regr: 0.0244 - detector_cls: 0.2291 - detector_regr: 0.0518No positive ROIs.
368/4579 [=>…] - ETA: 35:19 - rpn_cls: 0.2612 - rpn_regr: 0.0244 - detector_cls: 0.2285 - detector_regr: 0.0519No positive ROIs.
398/4579 [=>…] - ETA: 34:43 - rpn_cls: 0.2504 - rpn_regr: 0.0243 - detector_cls: 0.2188 - detector_regr: 0.0525No positive ROIs.
424/4579 [=>…] - ETA: 34:15 - rpn_cls: 0.2419 - rpn_regr: 0.0242 - detector_cls: 0.2114 - detector_regr: 0.0530No positive ROIs.
427/4579 [=>…] - ETA: 34:12 - rpn_cls: 0.2410 - rpn_regr: 0.0242 - detector_cls: 0.2106 - detector_regr: 0.0531No positive ROIs.
428/4579 [=>…] - ETA: 34:11 - rpn_cls: 0.2407 - rpn_regr: 0.0242 - detector_cls: 0.2104 - detector_regr: 0.0531No positive ROIs.
452/4579 [=>…] - ETA: 33:49 - rpn_cls: 0.2336 - rpn_regr: 0.0242 - detector_cls: 0.2042 - detector_regr: 0.0535No positive ROIs.
482/4579 [==>…] - ETA: 33:22 - rpn_cls: 0.2253 - rpn_regr: 0.0241 - detector_cls: 0.1972 - detector_regr: 0.0540No positive ROIs.
709/4579 [===>…] - ETA: 30:39 - rpn_cls: 0.1802 - rpn_regr: 0.0236 - detector_cls: 0.1610 - detector_regr: 0.0564No positive ROIs.

It means no positive RoI for training the RCNN in this image. Just for debug, user can feel free to ignore it. It is common to occur at the start of training. When the training converges, this will occur less frequently.

Reference: Faster RCNN ROI issue - #5 by Morganh

Thank you. I removed some images with small objects and that message disappeared.