dimension mismatch in test evaluation

[polo5]

The code to evaluate the test mAP fails due to a factor of 10 mismatch in the ground truth and model scores

Evaluating on test set:

File "src/train_cls.py", line 169, in <module>
    ap = metrics.average_precision_score(gts[i][gts[i]<=1], scr[i][gts[i]<=1]-1e-5*gts[i][gts[i]<=1])
IndexError: boolean index did not match indexed array along dimension 0; dimension is 49520 but corresponding boolean dimension is 4952

Evaluating on val set:

  File "src/train_cls.py", line 169, in <module>
    ap = metrics.average_precision_score(gts[i][gts[i]<=1], scr[i][gts[i]<=1]-1e-5*gts[i][gts[i]<=1])
IndexError: boolean index did not match indexed array along dimension 0; dimension is 25100 but corresponding boolean dimension is 2510

Here the ground truths are shape (20, X) but the scores are (20, X*10). I'm running on python 2.
My test_resize_layer.cpp failed in the 'make test' compiling so I removed it. Perhaps this is causing those shape mismatches now? Thanks for any tips.

[polo5]

After further investigation it seems that each row in r['fc8'] is equal, and keeping only the first row fixes the shape mismatch. This is done by changing the folowing line:

#scr.append(1 * r['fc8'])
scr.append(1*np.expand_dims(r['fc8'][0], axis=0)) # (10,20) -> (1,20)

For 1 crop only the code runs without errors but produces poor mAP results (order of 3%), which is much lower than the 53% you get by random guessing...

[polo5]

Nevermind, this is what happens if you give a (faulty?) deploy.prototxt as input as opposed to a train.prototxt. If you're looking for a working example have a look here: https://github.com/jeffdonahue/bigan

[jianghaojun]

@polo5 Did you solve the mismatch problem? I spent a lot of days debugging the code but failed. The jeffdonahue/bigan code has same problem. Besides, my test_resize_layer.cpp also failed in the 'make test' compiling.
Help,please.