Maskrcnn-benchmark: the Retinanet is not available right now?

[feihuidiqiu]

❓ Questions and Help

I test the Retinanet by replace the
config_file = "../configs/caffe2/e2e_mask_rcnn_R_50_FPN_1x_caffe2.yaml"
to
config_file = "../configs/retinanet/retinanet_R-50-FPN_1x.yaml"
in Mask_R-CNN_demo.ipynb,
and then i get 0 bbox in the result, is the Retinanet under develop or i need to change the input of picture, thank you for any help~

[fmassa]

Hi,

We haven't uploaded pre-trained models for RetinaNet.

@chengyangfu do you know if the pre-trained models from Caffe2 work with the version in here, or do we need to modify the c2_model_loading for that?

[feihuidiqiu]

Hi,

We haven't uploaded pre-trained models for RetinaNet.

@chengyangfu do you know if the pre-trained models from Caffe2 work with the version in here, or do we need to modify the c2_model_loading for that?

Thank you,
I didn't notice that.

[chengyangfu]

@fmassa
Unfortunately, we need to modify the model from Caffe2 when loading it. I will check that part this weekend.

@feihuidiqiu,
You can find some RetinaNet models trained on this repository here. https://github.com/facebookresearch/maskrcnn-benchmark/pull/102
You just need to change cfg.merge_from_list(["MODEL.DEVICE", "cpu"]) to
cfg.merge_from_list(["MODEL.DEVICE", "cpu", "MODEL.WEIGHT", "THE_MODEL_YOU_DOWNLOAD"])

[fmassa]

@chengyangfu you probably just need to rename it according to https://github.com/facebookresearch/maskrcnn-benchmark/blob/master/maskrcnn_benchmark/utils/c2_model_loading.py

[Computational-Camera]

Actually, after 36 hours' training with a single GPU, the results from Retinanet does not make me disappointed.

[zxhr2793]

Actually, after 36 hours' training with a single GPU, the results from Retinanet does not make me disappointed.

Which backbone and what are the results?