search

ijkguo / mx-rcnn

Parallel Faster R-CNN implementation with MXNet.
Other
671 stars 290 forks source link

how to print out each category's accuracy using faster rcnn? #80

Closed yian2271368 closed 6 years ago

yian2271368 commented 6 years ago

i am doing a object detection project for 9 classes(including background), but in the end it prints out the loss and classfication score for overall classes. so i am wondering how could I print out loss and acc(ie, rnploss&acc, rcnn loss&acc) for each class?

ijkguo commented 6 years ago

You might need to write metrics. See rcnn/core/metric.py for examples.

Ram-Godavarthi commented 6 years ago

@ijkguo I have done training on 2 classes.. I got this results after 5 epochs.

INFO:root:testing 197/200 data 0.0125s net 0.3531s post 0.0002s INFO:root:testing 198/200 data 0.0126s net 0.3365s post 0.0005s INFO:root:testing 199/200 data 0.0150s net 0.3315s post 0.0003s INFO:root:Writing pedestrian VOC results file INFO:root:Writing bicycle VOC results file INFO:root:VOC07 metric? Y INFO:root:AP for pedestrian = 0.8070 INFO:root:AP for bicycle = 0.8183 INFO:root:Mean AP = 0.8126

Now i want to use this model to predict the classes. should i use demo.py to visualize the results or should i do something else?

Actually i executed demo.py . Here i could feed only 1 image as input to check the results. Is there any other way where i can give some 100's of images and visualize the output on all images.?

When i do demo.py on single image. i am getting only 1 class(box) per image. i am not getting 2 boxes on single image. what is the reason for this??

Please provide me some solution for this.

ijkguo commented 6 years ago

Redirect to https://github.com/ijkguo/mx-rcnn/issues/95.