search

Cadene / murel.bootstrap.pytorch

MUREL (CVPR 2019), a multimodal relational reasoning module for VQA
https://arxiv.org/abs/1902.09487
BSD 3-Clause "New" or "Revised" License
195 stars 24 forks source link

Error while trying to train or validate #8

Closed archanray closed 5 years ago

archanray commented 5 years ago

Hi, thanks for the code.

I am trying to re-train MuRel. I got the following error: Object arrays cannot be loaded when allow_pickle=False

Attached is the output log. Please let me know if I'm doing something incorrectly.

Command: CUDA_VISIBLE_DEVICES=4,5,6,7 python -m bootstrap.run -o murel/options/vqa2/murel.yaml 2>&1 | tee murel_output.txt

murel_output.txt

Cadene commented 5 years ago

@LousyLory hi, thanks for your interest. I don't know what cause this issue.

Did you try to debug by adding import pdb;pdb.set_trace() in /home/ubuntu/anaconda3/envs/murel/lib/python3.7/site-packages/skipthoughts/skipthoughts.py (line 61)?

There is an issue when the pretrained skipthoughts parameters are loaded in numpy... :/

archanray commented 5 years ago

I fixed this by doing the following: in file /home/ubuntu/anaconda3/envs/murel/lib/python3.7/site-packages/skipthoughts/skipthoughts.py

line 61: original: params = numpy.load(path_params, encoding='latin1') modified: params = numpy.load(path_params, encoding='latin1', allow_pickle=True)

line 70: original: params = numpy.load(path_params, encoding='latin1') modified: params = numpy.load(path_params, encoding='latin1', allow_pickle=True)

I think the issue stems from a numpy version mismatch. I am using 1.16.4.