Yang

mark

sorry, I didn't convert the resnet34 model

Sorry for the keras converter, I can't find the code. There is no trick for me to convert a model for any source to caffe. I load the weight from...

@zhujiapeng You can find the download urls here https://github.com/soeaver/caffe-model/tree/master/cls. And the input size problem please check https://github.com/soeaver/caffe-model/issues/11

I think the complexity of the network after roipooling causes this phenomenon. The VGG16-faster-rcnn only has two fc layers after roipooling, but the ResNet50-faster-rcnn has the stage-4 (3 bottleneck =...

I think you should pay attention on image pre-processing: The RGB channel or BGR channel; The mean value and std value

Hi, you mean multi-scale training for semantic seg? Usually, ms training will lead to a slightly larger and unstable loss. I think you can look at the final result of...

Actually, I didn't do a single-scale training experiment. But as many papers say, ms training and random flipping will improve 1-3% mIoU as PASCAL VOC dataset.

yes, the groundtruth of ILSVRC2015 is different with ILSVRC2012

You can finetune the model on ILSVRC2012, but I don't think it's necessary, the images of ILSVRC2012 and ILSVRC2015 are exactly the same, only the order of the labels is...