DCH: low results

[hbellafkir]

I trained DCH on COCO using the default parameters in val_train script, which sets the iter_num to 2000. The results too low compared to the paper:

iter 33 prec 0.3673785145624646 rec 0.9312258941859837 mAP 0.4742411791076638
i2i_by_feature 0.5373119936770314
i2i_after_sign 0.48951766955994036
i2i_prec_radius_2 0.44337251780011877
i2i_recall_radius_2 0.6585969771414468
i2i_map_radius_2 0.518241127343017

Did i miss something the configuration file?

[fegler]

Did u find the missing configuration?
If you find, could you share your hyperparameter setting?
I also think there's something missing..
e.g., the default value of the quantization loss parameter(q_lambda) is set to 0.

[hbellafkir]

No i didn't. I'm still waiting for a answer.
Yes I noticed that too, q_lambda is set to 0 not only in DCH, but also in DTQ.

[sacralitus]

No i didn't. I'm still waiting for a answer.
Yes I noticed that too, q_lambda is set to 0 not only in DCH, but also in DTQ.

Hi! Could you help me please? I'm beginner in DL and have no idea how to test model I have trained. Model is somehow saved in .npy format. Don't know how to work in such a format therefore can't put bunch of my test images to test the CBIR model :(