Py-faster-rcnn: Configuration yml not being passed down to lib/fast_rcnn/test.py

[alantrrs]

Currently if you run the script ./experiments/scripts/faster_rcnn_alt_opt.sh [GPU_ID] [NET] it fails during testing mode with: Input blob arguments do not match net inputs

I traced that back to HAS_RPN = False in lib/fast_rcnn/config.py even though the experiments/cfgs/faster_rcnn_alt_opt.yml sets it to True.

I believe the .yml file should override the default settings given by config.py, is that correct?

[rbgirshick]

I'm unable to replicate this issue.

[debidatta]

I think this is an easy_dict issue. I am facing a similar problem where I try to set the config using the experiments/cfgs/faster_rcnn_alt_opt.yml.

To investigate the problem, I changed the train_net script as follows:

if args.cfg_file is not None:
        cfg_from_file(args.cfg_file)
        print cfg.TRAIN['PROPOSAL_METHOD']
        print cfg.TRAIN.PROPOSAL_METHOD

This gives an output

selective_search

I suspect the _merge_a_into_b(a, b) in config.py creates the ambiguity.

[EthanZhu90]

I fixed the issue by:
cfg_from_file(cfg_file)
cfg = edict(cfg) ## add this line

[fanw52]

hello , whether the PROPOSE_METHOD is the 'gt' or not ,the 'selective search 'method will not be executed,that is to say ,In train phase, the code

PROPOSAL_METHOD:gt

can be overlooked.
and I can't find PROPOSAL_METHOD:gt the in the test phase @rbgirshick @debidatta @EthanZhu90