alt training error about fpn HOT 2 OPEN

I modified the function train_fast_rcnn(self, sess, vs_names, max_iter, rpn_file, init_model=None):
feed_dict={
self.net.data: blobs['data'],
self.net.leveled_rois_0: blobs['leveled_rois_0'],
self.net.leveled_rois_1: blobs['leveled_rois_1'],
self.net.leveled_rois_2: blobs['leveled_rois_2'],
self.net.leveled_rois_3: blobs['leveled_rois_3'],
self.net.leveled_rois_4: blobs['leveled_rois_4'],
......
by adding self.net.leveled_rois_4: blobs['leveled_rois_4'], the error disappeared,but the the Mean AP = 0.1282, a very poor result compared to end-to-end training. I trained the network using VOC2007.

from fpn.

@missliontheway Hi, I also have this problem, have you figured it out? If so, would you please share it? I have stuck in this code a few days. Thank you so much.

from fpn.