search

Xilinx / Vitis-AI-Tutorials

MIT License
360 stars 145 forks source link

Questions about "Test net output #0: detection_eval = 0" #5

Open keson1984 opened 4 years ago

keson1984 commented 4 years ago

Hello, I download the Vitis-AI-ssd/SSD and retrain the Mobilenetv2-SSD in the Vitis-AI-ssd/SSD/workspace/Mobilenetv2-SSD. the log as follow:

I0521 00:17:13.147367 1523 solver.cpp:772] Iteration 6000, Testing net (#0) I0521 00:17:13.148303 1523 net.cpp:743] Ignoring source layer mbox_loss I0521 00:19:13.689599 1523 solver.cpp:885] Test net output #0: detection_eval = 0 I0521 00:19:19.519392 1523 solver.cpp:270] Iteration 6000 (0.149189 iter/s, 670.289s/100 iter), loss = 3.38898, remaining 212 hours and 14 minutes I0521 00:19:19.519454 1523 solver.cpp:291] Train net output #0: mbox_loss = 3.49844 (* 1 = 3.49844 loss) I0521 00:19:19.519470 1523 sgd_solver.cpp:106] Iteration 6000, lr = 0.001 ...........................................

why "Test net output #0: detection_eval = 0". it confuse me for almost all the day, I can't find any useful solution in google. ( I use caffe-xilinx 1.1 )

train.sh: /workspace/caffe-xilinx/build/tools/caffe train -solver="/workspace/Vitis-AI-ssd/SSD/workspace/Mobilenetv2-SSD/solver.prototxt" \ -weights="/workspace/Vitis-AI-ssd/SSD/workspace/Mobilenetv2-SSD/pretrained.caffemodel" -gpu 0,1,2,3 2>&1 | tee train.log:q

luxiaojian77 commented 3 years ago

Hello. Did you solve it? I suffered the same problem.

DrChungAlbert commented 3 years ago

This tutorial, however, uses Caffe provided within Vitis AI 1.1. I suffered the same problem @ Vitis AI master , change it to Vitis AI 1.1 and got the correct results.

sunshine411 commented 3 years ago

@DrChungAlbert I use Vitis-AI-1.2/tree/master/AI-Model-Zoo/caffe-xilinx suffered the same problem Test net output #0: detection_eval = 0 also test on xilinx/vitis-ai:lastest get same problem.

I don't understand which version you are using to solve this problem。 Can you tell me more specific information？