issues
search
mertyg
/
vision-language-models-are-bows
Experiments and data for the paper "When and why vision-language models behave like bags-of-words, and what to do about it?" Oral @ ICLR 2023
MIT License
261
stars
15
forks
source link
issues
Newest
Newest
Most commented
Recently updated
Oldest
Least commented
Least recently updated
Performance Recall@1 COCO (table 6)
#40
FiorenzoParascandolo1
opened
1 month ago
0
Does NegCLIP now support mulit-gpu and grad_accumulation？
#39
Vicent0205
opened
5 months ago
0
A question on results from Figure 2 and bag-or-wordness
#38
iburenko
opened
8 months ago
0
Why does `visual_genome_relation.json` still contain symmetric relations?
#37
lcxrocks
closed
9 months ago
4
Question regarding numbers in Figure 1
#36
YunYunY
closed
9 months ago
1
Evaluation bug when using GELU vs QuickGELU -- changes the results for some benchmarks
#35
bryant1410
opened
11 months ago
1
Could you share the fine-tuned fiber checkpoint with EqSim constraints?
#34
ytaek-oh
closed
11 months ago
1
Questions on evaluation results
#33
ytaek-oh
opened
11 months ago
1
about the performance of originial CLIP
#32
hiker-lw
closed
9 months ago
16
How you run expreriments with batch size 1024 on a Single RTX-2080Ti
#31
wujianP
closed
1 year ago
1
I cannot run on RTX 3060 with batch-size=256!
#30
shuguang99
closed
1 year ago
2
I can't reproduce Table 6
#29
shuguang99
closed
1 year ago
2
parameter file problem
#28
haoshuai714
closed
1 year ago
2
train negCLIP result problem
#27
haoshuai714
closed
1 year ago
10
Similarity scores for NegCLIP are pretty similar
#26
kochsebastian
closed
1 year ago
0
Model weights of regular COCO finetuning.
#25
wildphoton
opened
1 year ago
0
Flava image preprocessing
#24
DianeBouchacourt
closed
1 year ago
8
slow evaluation for xvlm
#23
lezhang7
closed
1 year ago
1
Projections W_i and W_t
#22
DianeBouchacourt
closed
1 year ago
7
Where to find the training data of NegCLIP?
#21
yu-wyatt-wu
closed
1 year ago
2
dataset size of flickr and coco order datasets
#20
HarmanDotpy
closed
1 year ago
4
matrix size for contrastive learning in model training
#19
Lycus99
closed
1 year ago
5
Table 6, COCO and Flickr Image/Text R@1 results
#18
HarmanDotpy
closed
1 year ago
8
Calling model.eval() when computing scores otherwise non-deterministic results (torch._no_grad_() is not enough)
#17
DianeBouchacourt
closed
1 year ago
1
Requirements (e.g. torch versions)
#16
DianeBouchacourt
closed
1 year ago
4
Questions on BLIP score computation
#15
DianeBouchacourt
closed
1 year ago
3
Fix a bit typo README
#14
guspan-tanadi
opened
1 year ago
0
Code for generating negatives for training NegCLIP
#13
HarmanDotpy
closed
1 year ago
2
fast
#12
vinid
closed
1 year ago
0
eval coco order and flickr order
#11
lezhang7
closed
1 year ago
3
Concrete benchmark results of attributes understanding
#10
Yangyi-Chen
closed
1 year ago
5
mismatching results on compositional task
#9
lezhang7
closed
1 year ago
25
question about VG-Relation categories
#8
hiker-lw
closed
1 year ago
4
Models are not in eval() mode.
#7
linzhiqiu
closed
1 year ago
1
Will the CoCo-order and Flickr-order dataset be released?
#6
linzhiqiu
closed
1 year ago
2
why concat df to all_df?
#5
lezhang7
closed
1 year ago
1
Exact hyperparameters for NegCLIP training. and question about imagenet accuracy reported in the paper
#4
HarmanDotpy
closed
1 year ago
20
Can you release the dataset during the finetuning of NegCLIP? Thanks!
#3
mu-cai
closed
1 year ago
5
When can you provide code and dataset？
#2
BigHyf
closed
1 year ago
5
Thanks for your great work! When are you planning to share the code?
#1
mu-cai
closed
1 year ago
6