Training the RFCN with Selective search boxes

[SilviaLauraPintea]

Hi,

I am trying to run the network training on Pascal-VOC without RPN boxes, but with selective search boxes. So I have edited the configuration file:

1 EXP_DIR: rfcn_end2end 2 TRAIN: 3 HAS_RPN: False 4 IMS_PER_BATCH: 1 5 BBOX_NORMALIZE_TARGETS_PRECOMPUTED: True 6 BBOX_REG: False 7 PROPOSAL_METHOD: selective_search 8 BG_THRESH_LO: 0.1 9 BATCH_SIZE: 128 10 AGNOSTIC: True 11 SNAPSHOT_ITERS: 10000 12 TEST: 13 HAS_RPN: False 14 BBOX_REG: False 15 AGNOSTIC: True

And I have downloaded the pretrained imagenet models and the selective search boxes but when I try to run it I get the error:

Solving... /var/scratch/spintea/Repositories/ms-caffe/tools/../lib/roi_data_layer/minibatch.py:102: VisibleDeprecationWarning: using a non-integer number instead of an integer will result in an error in the future fg_inds, size=fg_rois_per_this_image, replace=False) /var/scratch/spintea/Repositories/ms-caffe/tools/../lib/roi_data_layer/minibatch.py:115: VisibleDeprecationWarning: using a non-integer number instead of an integer will result in an error in the future bg_inds, size=bg_rois_per_this_image, replace=False) /var/scratch/spintea/Repositories/ms-caffe/tools/../lib/roi_data_layer/minibatch.py:122: VisibleDeprecationWarning: using a non-integer number instead of an integer will result in an error in the future labels[fg_rois_per_this_image:] = 0 Traceback (most recent call last): File "tools/train_net.py", line 112, in max_iters=args.max_iters) File "/var/scratch/spintea/Repositories/ms-caffe/tools/../lib/fast_rcnn/train.py", line 204, in train_net model_paths = sw.train_model(max_iters) File "/var/scratch/spintea/Repositories/ms-caffe/tools/../lib/fast_rcnn/train.py", line 145, in train_model self.solver.step(1) File "/var/scratch/spintea/Repositories/ms-caffe/tools/../lib/roi_data_layer/layer.py", line 145, in forward blobs = self._get_next_minibatch() File "/var/scratch/spintea/Repositories/ms-caffe/tools/../lib/roi_data_layer/layer.py", line 63, in _get_next_minibatch return get_minibatch(minibatch_db, self._num_classes) File "/var/scratch/spintea/Repositories/ms-caffe/tools/../lib/roi_data_layer/minibatch.py", line 56, in get_minibatch num_classes) File "/var/scratch/spintea/Repositories/ms-caffe/tools/../lib/roi_data_layer/minibatch.py", line 127, in _sample_rois roidb['bbox_targets'][keep_inds, :], num_classes) KeyError: 'bbox_targets'

I am not sure what I am doing wrong. Could you maybe give an example of how to run the network over input selective search boxes (or other boxes) instead of RPN boxes?

Thanks a lot! Cheers, Silvia