muye5

thanks for your great job. I found it didn't work for large scale model with huge sparse embedding parameters. for the apply_sparse function may be not very efficient as AdamOptimizer...

hyperparameter & pre-train issues

3

- Does this detection model train totally from scratch with COCO-Dataset or VOC? Should we do pre-train with ImageNet? - GRID_W = 18 & GRID_H = 10 How to set...

I think it's a little strange about the way of roi_w_scale settle (same with roi_h_scale): roi[1] / anchor[1]. Is the roi[1] the ratio of bbox width with original image width...

inference bug?

2

phi never be used in inference?

ROLO VS YOLO？

1

If detection is needed to do for every frame, what's the point of rolo. For I think only yolo enough do predict the bounding box, with simple iou, we can...