Thanks a lot for your code.But the deploy.prototxt seems wrong.Can you upload the right one?

Thanks you for great work. I meet some problems. When I test with your mobilenet-yolo such as ./build/examples/ssd/ssd_detect models/Pelee/deploy.prototxt models/Pelee/pelee_YOLO_iter_51000.caffemodel -file_type image -wait_time 3000 -mean_value 0.5,0.5,0.5 -normalize_value 0.007843 -confidence_threshold 0.3
F1014 19:03:47.823598 48275 insert_splits.cpp:29] Unknown bottom blob 'label' (layer 'detection_eval', bottom index 1)
*** Check failure stack trace: ***
@ 0x7fc78cdacdaa (unknown)
@ 0x7fc78cdacce4 (unknown)
@ 0x7fc78cdac6e6 (unknown)
@ 0x7fc78cdaf687 (unknown)
@ 0x7fc78d8b3c2a caffe::InsertSplits()
@ 0x7fc78d81bffa caffe::Net<>::Init()
@ 0x7fc78d81b4c7 caffe::Net<>::Net()
@ 0x40e07a Detector::Detector()
@ 0x410749 main
@ 0x7fc78a70cf45 (unknown)
@ 0x40de19 (unknown)
@ (nil) (unknown)
could you give me some advice? thank you

How to use your trained .caffemodel as a pretrained weights (transfer learning) in the model which we train. Is it possible in caffe ?

I want to run your model on an Android phone. So I need to turn the caffe model into tesorflow's .pb file. I use code (https://github.com/ethereon/caffe-tensorflow) to transform the caffe model, but some mistake happend:Message type "caffe.LayerParameter" has no field named "yolov3_detection_output_param". I guess its because you write the function yourself. So would you please help me to turn the pva_yolo model into a .pb file? thanks a lot.