DG-Net
DG-Net copied to clipboard
NVlabs
I got so low rank1
Firstly, thanks for the work and sorry for my low knowledge. i down the teach model and train the DGnet 100000 Times，but the source was so low ，has any met it and do you give me some suggest
Hi @jinyl777 , Thanks for your attention on our work. How about your results? Do you conduct the traditional re-ID evaluation or transfer learning between the datasets? Could you show me more details? So I could help you to debug the result.
I am honored that you can reply to me, I download the teach model according to the readme, then python train.py --config configs/latest.yaml then test and envlution but got zhe rank1 is 0.001 and then if iwant to train my own data ，what should i do . if you can teach me i weill be honored
@jinyl777 First of all, would you like to evaluate the trained model? Does the trained model achieve the accurate performance?
yes，frist i want trian in my computer and evaluate it , then trian myself dataset
@jinyl777 I mean you may directly use my trained model first. https://github.com/NVlabs/DG-Net#testing
thank you very match , i use this modle i got the nice sorce . if i want to use it to my detaset ，do I have to train my own teach model
Okay. Could you show me your loss curves or some logs? Does the generated image look good?
Because I am still recreating your results, so my data set has not been tested, I will tell you after the test. thanks very math
Hi, I used your code to train the generated image to work well, but loss is not converging. Epoch: 12 Iteration: 00083796/00100000 DLoss: 2.9820 Reg: 0.0055 L_total: 54.3989, L_gan: 3.0339, Lx: 0.7561, Lxp: 0.7570, Lrecycle:0.3373, Lf: 0.9595, Ls: 0.2201, Recon-id: 7.1530, id: 20.2080, pid:14.4382, teacher: 6.5358 Elapsed time in update: 2.197724
HI @jinyl777 , You may modify the re-id learning rate. Try small learning rate.