Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

what is checkpoint-29-66420 #191

Open
Vincent-Ww opened this issue Apr 3, 2022 · 2 comments
Open

what is checkpoint-29-66420 #191

Vincent-Ww opened this issue Apr 3, 2022 · 2 comments

Comments

@Vincent-Ww
Copy link

From this page https://github.com/microsoft/Oscar/blob/master/MODEL_ZOO.md, For Oscar on image captioning task, there is only one checkpoint model: checkpoint-29-66420.
This checkpoint model is trained by only cross-entropy or it's finetuned by cross-entropy and CIDER optim?

When I run the inference script with checkpoint-29-66420 on MSCOCO, I got the results:
{'Bleu_1': 0.7559338836672688, 'Bleu_2': 0.6008669959375728, 'Bleu_3': 0.46893216244997465, 'Bleu_4': 0.3658244160093896, 'METEOR': 0.3040389540024913, 'ROUGE_L': 0.5856375658366109, 'CIDEr': 1.2412284516798686, 'SPICE': 0.231772161443854}

So I guess checkpoint-29-66420 model is only pre-trained by cross-entropy but without CIDEr optim, is that correct?

@ignasa007
Copy link

I think only CE. I am replicating the results and get the same checkpoint after CE optimisation. Further, the authors specify 30 epochs for CE optimisation, so should be it.

@Nidadadadada
Copy link

Nidadadadada commented Nov 13, 2022

@Vincent-Ww @ignasa007 Hi, I run the inference script with checkpoint-29-66420 and the result is weird, do you meet the same problem? I download the checkpoint 29-66420 using the same link. Thanks!
微信图片_20221113201654
微信图片_20221113201659
This is the command line I wrote:
python oscar/run_captioning_official.py --do_test --do_eval --test_yaml test.yaml --per_gpu_eval_batch_size 64 --num_beam 5 --max_gen_length 20 --eval_model_dir /home/skn/Oscar-master/output_ceshi/checkpoint-29-66420

Looking forward to your reply!

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

3 participants