tf-faster-rcnn A simple idea to solve the error : "rpn_loss

There is an error with bounding boxes' coordinates,when some of coordinate values become 65535.

So, we can open the file named "minibatch.py" , find the line:

gt_boxes = np.empty((len(gt_inds), 5), dtype=np.float32)

and insert some codes in front of it :

boxes = roidb[0]['boxes'][gt_inds, :]         
for idx in range(len(boxes)):
  &nbsp; boxes[idx][boxes[idx]==65535] = 0

这可能算不上是个好的解决方案 ~

Jan 16 '18 08:01 yaoing

the underlying reason might be your dataset is 0-based rather than 1-based in coordinates?

Jan 18 '18 00:01 endernewton

I faced that issue when forget about -1 for x_max and y_max in dataset loader

Mar 26 '18 09:03 chichivica

I also get this issue. And I have deleted the -1. But it still happen! Does it have a better solution now?

Jun 20 '18 07:06 FMsunyh

Before running clean your annotation files by checking these conditions. if int(ymin) > int(height): if int(ymax) > int(height): if int(xmin) >= int(xmax): if int(ymin) >= int(ymax): if int(xmin) > int(width): if int(xmax) > int(width) : if int(xmin) <=0 if int(ymin) <=0 if int(xmax) <=0: if int(ymax) <=0:

Jun 20 '18 09:06 rnsandeep

@FMsunyh Did you delete the .pkl files created in in the cache folder of the dataset? For me deleting the -1 didn't have any effect too at first. But deleting the -1 and after that the .pkl files solved it for me.

Jun 25 '18 02:06 kinzzoku

@kinzzoku yes, I've deleted .pkl file and rebuild it again, but same issue with me.

Jun 25 '18 06:06 FMsunyh

There is an error with bounding boxes' coordinates,when some of coordinate values become 65535.

So, we can open the file named "minibatch.py" , find the line:
gt_boxes = np.empty((len(gt_inds), 5), dtype=np.float32)
and insert some codes in front of it :
boxes = roidb[0]['boxes'][gt_inds, :]         
for idx in range(len(boxes)):
  &nbsp; boxes[idx][boxes[idx]==65535] = 0
这可能算不上是个好的解决方案 ~ boxes[idx][boxes[idx]==65535] = 0 请问这样做是为什么？