arch_cfg.yaml not found

[SoftwareApe]

infer.py tries to load arch_cfg.yaml and data_cfg.yaml. Neither file can be found.

    # open arch config file
    try:
        print("Opening arch config file from %s" % FLAGS.model)
        ARCH = yaml.safe_load(open(FLAGS.model + "/arch_cfg.yaml", 'r'))
    except Exception as e:
        print(e)
        print("Error opening arch yaml file.")
        quit()

    # open data config file
    try:
        print("Opening data config file from %s" % FLAGS.model)
        DATA = yaml.safe_load(open(FLAGS.model + "/data_cfg.yaml", 'r'))
    except Exception as e:
        print(e)
        print("Error opening data yaml file.")
        quit()

[sebaleme]

We can retrain the model to generate them, but then providing the model alone is not enough

[TiagoCortinhal]

@sebaleme I will upload them later today but you can use the ones generated by retraining (with standard options) as it will be the same.

[sebaleme]

Hi, The arch_conf is in the repo, but not at the expected position: it is directly at the root level, with the following name: salsanext.yml One question: in SalsaNet, you say that your network is independent of the projection type, and that BEV or SFV yields the same results. In SalsaNext however, you only consider SFV, and don t speak about BEV anymore. Is it the KNN postprocessing module which forces you to use SFV, or you could have also reuse BEV?

[TiagoCortinhal]

@sebaleme, kNN should work in both types of projections, but for the scope of this paper we focused on SFV.