Thank you once again for making the code public and pretty easy to understand with such a detailed paper and instructions.
I was trying to generate qualitative examples / visualization by plotting the final RNCDL model outputs by adding the plotting logic in this file: ./discovery/evaluation/evaluator_discovery.py, in the process function of DiscoveryEvaluatorclass.
Since the outputs from here are directly used for evaluation/calculation of mAP, I'm using the outputs variable to try and plot the model outputs.
However the visualization of the examples is quite noisy and there seems to be lot of false positives. Please look at these images:
Am i doing something wrong? could you please check? Im using the exact same settings for all the hyperparameters/confidence threshold which is used for the evaluation purposes.
Hello,
Thank you once again for making the code public and pretty easy to understand with such a detailed paper and instructions.
I was trying to generate qualitative examples / visualization by plotting the final RNCDL model outputs by adding the plotting logic in this file:
./discovery/evaluation/evaluator_discovery.py
, in theprocess
function ofDiscoveryEvaluator
class.Since the outputs from here are directly used for evaluation/calculation of mAP, I'm using the
outputs
variable to try and plot the model outputs.However the visualization of the examples is quite noisy and there seems to be lot of false positives. Please look at these images:
Am i doing something wrong? could you please check? Im using the exact same settings for all the hyperparameters/confidence threshold which is used for the evaluation purposes.