phellonchen / dmrm Goto Github PK

Hello, all the data I downloaded are H5 files, and they are very large. How should I deal with them?

Traceback (most recent call last):
File "main_v1.0.py", line 69, in
eval_dset = dl.validate(input_img_h5=args.input_img_val_h5, input_ques_h5=args.input_ques_h5,
File "/home/zutnlp/0_jipeiyu/DMRM-master-v1/misc/dataLoader_v1.py", line 197, in init
self.hdf_reader = ImageFeaturesHdfReader(
File "/home/zutnlp/0_jipeiyu/DMRM-master-v1/misc/readers.py", line 293, in init
with h5py.File(self.features_hdfpath, "r") as features_hdf:
File "/home/zutnlp/anaconda3/lib/python3.8/site-packages/h5py/_hl/files.py", line 406, in init
fid = make_fid(name, mode, userblock_size,
File "/home/zutnlp/anaconda3/lib/python3.8/site-packages/h5py/_hl/files.py", line 173, in make_fid
fid = h5f.open(name, flags, fapl=fapl)
File "h5py/_objects.pyx", line 54, in h5py._objects.with_phil.wrapper
File "h5py/_objects.pyx", line 55, in h5py._objects.with_phil.wrapper
File "h5py/h5f.pyx", line 88, in h5py.h5f.open
OSError: Unable to open file (file signature not found)

Could you help me find out what the problem is? Thank you very much!

Hello, I found input_imgid in evut_1.0.py, which is features_faster_rcnn_x101_train_v1.0_imgid.json, you can provide it， thank you

I clone the code from source for training. And I found the result in v1.0 were bad.

Epoch 12: mrr: 0.425225 R1: 0.326793 R5 0.512984 R10 0.560126 Mean 23.555378 time: 112.83
Epoch 13: mrr: 0.425715 R1: 0.326793 R5 0.514390 R10 0.561240 Mean 23.428488 time: 117.13
Epoch 14: mrr: 0.425800 R1: 0.326502 R5 0.514583 R10 0.561531 Mean 23.346754 time: 117.97

then the model stop training.

how to get the same result in your paper

I want to train with my own dataset, .h5 file I should use faster-rcnn that layer, is it after the roi, but didn't 8000x26)

File "main_v1.0.py", line 90, in
model = train(model, train_loader, eval_loader, args)
File "/home/bigdata2/xnq/DMRM-master/train.py", line 67, in train
train_data = next(train_iter)
IndexError: arrays used as indices must be of integer (or boolean) type

Are the three FastRCNN *.train.h5, val.h5 and test.h5 feature files extracted on VisDial v1.0?
Thanks for your attention.