transformer_nsc

[bai-24]

你好，我在运行transformer_nsc.yml时，碰到以下问题，请问怎么解决？
Hugginface transformers not installed; please visit https://github.com/huggingface/transformers
meshed-memory-transformer not installed; please run pip install git+https://github.com/ruotianluo/meshed-memory-transformer.git
Warning: coco-caption not available

[ruotianluo]

无视就可以了

[bai-24]

好的， 感谢，请问要训练transformer+nsc的话，直接运行transformer_nsc.yml就可以了吗？还是有什么先后顺序？

[ruotianluo]

应该还是要先用transformer.yml，然后再到nsc

[bai-24]

你好，请问用transformer.yml训练完后，不小心关了界面，Bleu、Meteor、Rouge等评价指标在哪里可以看见呢

[ruotianluo]

Tensorboard

[bai-24]

请问在生成图像描述时，出现以下错误，有什么解决办法吗？
Traceback (most recent call last):
File "tools/eval.py", line 77, in
lang_stats = eval_utils.language_eval(opt.input_json, predictions, n_predictions, vars(opt), opt.split)
File "g:\self-critical.pytorch\captioning\utils\eval_utils.py", line 80, in language_eval
mean_perplexity = sum([_['perplexity'] for _ in preds_filt]) / len(preds_filt)
ZeroDivisionError: division by zero

[ruotianluo]

command是什么

[bai-24]

输入为：python tools/eval.py --model log_fc_nsc/model-best.pth --infos_path log_fc_nsc/infos_fc_nsc-best.pkl --image_folder blah --num_images 10

[ruotianluo]

大多数pretrained 模型不支持任意图片的captioning，如果你想试任意图片，可以用readme里面的colab

[bai-24]

您好，请问bottomup feature训练好的transformer模型，如何测试COCO数据集中的图像描述效果 ***@***.*** From: Ruotian(RT) Luo Date: 2022-11-22 15:07 To: ruotianluo/self-critical.pytorch CC: bai-24; Author Subject: Re: [ruotianluo/self-critical.pytorch] transformer_nsc (Issue #277) 大多数pretrained 模型不支持任意图片的captioning，如果你想试任意图片，可以用readme里面的colab ― Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: ***@***.***>

[ruotianluo]

https://github.com/ruotianluo/self-critical.pytorch#evaluate-on-karpathys-test-split

[bai-24]

请问transformer训练时，可以使用Resnet101 feature吗？

[ruotianluo]

没试过，应该是可以的。因为resnet101比较长，训练比较慢我后来都不用了Ruotian LuoOn Jan 16, 2023, at 9:08 AM, bai-24 ***@***.***> wrote: 请问transformer训练时，可以使用Resnet101 feature吗？ —Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you commented.Message ID: ***@***.***>