DAMO-NLP-SG VCD issues - Githubissues

DAMO-NLP-SG / VCD

[CVPR 2024 Highlight] Mitigating Object Hallucinations in Large Vision-Language Models through Visual Contrastive Decoding

Apache License 2.0

195 stars 9 forks source link

issues

Newest

Newest Most commented Recently updated Oldest Least commented Least recently updated

Inference Sample

#23 Shinyzenith opened 1 week ago
1
Update vcd_sample.py to adapt to the latest transformers library

#22 MrChenFeng closed 1 week ago
0
About the experiment results in the Table.1

#21 tbbbk opened 3 weeks ago
2
Request for Parameter and Prompt Configurations Used in Table 3

#20 ShanWang-Shan closed 1 week ago
3
VCD in LVLMs

#19 sunzjz opened 1 month ago
0
About greedy search in VCD

#18 haohaodw opened 1 month ago
0
question about evaluation.

#17 tt6746690 opened 2 months ago
1
Inquiry Regarding Discrepancy in Results Using the Same Model and Methodology as Presented in Your Paper

#16 obananas opened 2 months ago
14
About the Experimental Setup

#15 haohaodw closed 2 months ago
4
Issues about importing llava

#14 Melon-Xu closed 2 months ago
0
`model_kwargs_cd` is always a copy from `model_kwargs`, is it a bug?

#13 teamwong111 closed 2 months ago
11
might be a waste of resources

#12 SDaoer closed 3 months ago
1
unable to reproduce the results of Table 7

#11 XueJiang16 closed 3 months ago
1
Multinomial Sampling May Not Ensure the Reliability of the Answer

#10 Eason-Qin closed 3 months ago
1
unable to reproducing the results of llava

#9 frankRenlf closed 3 months ago
10
Unable to reproduce results from greedy search

#8 qychen2001 closed 3 months ago
4
model_kwargs_cd and model_kwargs problem

#7 Mr-xiu closed 3 months ago
3
If two confidence levels of original and distorted inputs are high and similar, can plausibility constraints make negative effect?

#6 QiushiYang opened 6 months ago
0
Special Hyperparameter value or Checkpoints are needed?

#5 yejipark-m closed 6 months ago
2
Decoding hyperparameter details

#4 yuezih closed 8 months ago
6
can you provide the image subsets you used for evaluation because the whole set of gqa and coco is so large

#3 yfzhang114 closed 9 months ago
1
what about directly ablate the image or just use a BLACK image?

#2 DingchenYang99 closed 9 months ago
2
probability tensor contains either `inf`, `nan` or element < 0

#1 zli999 closed 10 months ago
8