Oscar

Oscar copied to clipboard

Dear scholar, I want to ask whether your elegant code includes the function about produce a description about the attribute and object for the designated bounding box. In your tools/demo_image.py...

alice-cool

Thanks for your great work! In Oscar paper, you mentioned beam size = 5 during inference. However, in your CIDEr optimization finetune command, you use beam size = 1. Is...

LeeYN-43

Cannot replicate VinVL VQA results

3

I was fine-tuning on VQA using VinVL features using the given scripts. However I am getting 74.82 evaluation accuracy, which is 1.3 lower then the reported one (76.12). It would...

Lizw14

Question on script code

1

Hey guys, i just want to figure out what's the details of the Script. if anyone can provide me the code, i'll be truly thanks. i mean this two parts...

pleasurepants

hi, Could I ask how to generate the downstream tasks' data provided from the link (https://biglmdiag.blob.core.windows.net/vinvl/datasets/TASK_NAME)? TASK_NAME could be coco_caption, nocaps, coco_ir, vqa, gqa, nlvr2.

ckmstydy

I want to see how this model performs on a significantly smaller dataset such as Flickr8k. Will I be able to train this model (from scratch) on Flickr8k?

mikkkeldp

File "./oscar/datasets/oscar_tsv.py", line 114, in __init__ img_feat_offset_map = self.img_feat_offset_map[dataset_name][chunk_id] KeyError: '0'

1

- INFO - root - loading lineidx: /userhome/Oscar/vinvl/pretrain_corpus/X152C4_frcnnbig2_exp168model_0060000model.roi_heads.nm_filter_2_model.roi_heads.score_thresh_0.2/gqa/QA_fileB.lineidx 07/20/2021 10:21:44 - INFO - root - loading lineidx: /userhome/Oscar/vinvl/pretrain_corpus/X152C4_frcnnbig2_exp168model_0060000model.roi_heads.nm_filter_2_model.roi_heads.score_thresh_0.2/gqa/predictions_gt.lineidx 39%|███████████████████████████████████████████████████████████████ | 3691028/9357057 [17:52

wangxiao5791509

Hi. I would like to ask regarding the att_masks for Image Captioning: In the data loading, you already prepare the att_masks : https://github.com/microsoft/Oscar/blob/master/oscar/run_captioning.py#L324 During inference, you re-process the att_masks here::...

homelifes

VQA finetune

2

Hi Oscar teams: I read your code and find that some parser_argument may loss, like "--model_name_or_path vinvl/model_ckpts/vqa/base/checkpoint-2000000" and "--tokenizer_name". Can you provide this file? Thanks.

1144181135

NoCaps VinVL test feature

1

Hello~ Thanks for uploading VinVL feature for NoCaps validation images. To generate results on test set, VinVL features for NoCaps test images are needed. Would you mind releasing VinVL features...

ChenYutongTHU