WeproposeanetworkforCongestedSceneRecognitioncalledCSRNettoprovideadata-drivenanddeeplearningmethodthatcanunderstandhighlycongestedscenesandperformaccuratecountestimationaswellaspresenthigh-qualitydensitymaps.TheproposedCSRNetiscomposedoftwomajorcomponents:aconvolutionalneuralnetwork(