search

yangxue0827 / R2CNN_FPN_Tensorflow

R2CNN: Rotational Region CNN Based on FPN (Tensorflow)
419 stars 139 forks source link

Configure parameters #70

Closed FoilHao closed 5 years ago

FoilHao commented 5 years ago

When you say configuring parameters in $R2CNN_ROOT/libs/configs/cfgs.py, do I really need to adjust the parameters? I mean, if they are well set, I might just leave them. I am a bit new in this area and working on detecting different kinds of bills and things like driving licenses, ID cards, bank cards and so on. I wonder if you can give me some advice on parameters, thank you in advance!

yangxue0827 commented 5 years ago

You mat need to modify ANCHOR_SCALES, ANCHOR_RATIOS, LR, DECAY_STEP and so on， it is determined by the characteristics of your data set. PS: I recommend using this repo, which has better performance.

FoilHao commented 5 years ago

Thanks very much for your advice

FoilHao commented 5 years ago

Meanwhile ,I wonder if this model works fine on detecting things like tickets, ID cards, bank cards,driving licenses and son on.

yangxue0827 commented 5 years ago

I have not tried these, so I am not quite sure. @FoilHao

FoilHao commented 5 years ago

Thank you, by the way, where should I put the test images before running test1.py?

FoilHao commented 5 years ago

I just tested with my own dataset and it worked pretty good! Thanks so much for your repo and it's super amazing! But I have one more question, I mean, I have had so many these days:) (Thanks for your patience) I used the VOCtestset(including images and annotations) to generate a tfrecord file, which was used in test1.py to output processed images, and the results look great. But in real situation, we usually have some images and want the model to directly output detection results from them. We don't want to label and convert them to tfrecords. So can I achieve this funtion(input images, run some codes and get results) through your work? Thanks so much!!!(This is probably my final question...:)

yangxue0827 commented 5 years ago

refer to https://github.com/DetectionTeamUCAS/FPN_Tensorflow/blob/master/tools/train_with_placeholder.py @FoilHao

FoilHao commented 5 years ago

Thanks!!!Such a great help!

refer to https://github.com/DetectionTeamUCAS/FPN_Tensorflow/blob/master/tools/train_with_placeholder.py @FoilHao