the training data of Meta fintuning how to extract?

[zhanggongjie]

the training data of Meta fintuning how to extract?

[ZhangGongjie]

Sorry, I didn't understand your question. You may use Chinese to describe your question and I can translate it into English so that everyone else would still understand.

Sorry for late reply, have a nice day!

[SamXiaosheng]

请问训练好了，要怎么推理？可以有inferrence文件吗

…

------------------ 原始邮件 ------------------ 发件人: "Gongjie ***@***.***>; 发送时间: 2022年4月18日(星期一) 下午3:25 收件人: ***@***.***>; 抄送: "生second ***@***.***>; ***@***.***>; 主题: Re: [ZhangGongjie/Meta-DETR] the training data of Meta fintuning how to extract? (Issue #19) Sorry, I didn't understand your question. You may use Chinese to describe your question and I can translate it into English so that everyone else would still understand. Sorry for late reply, have a nice day! — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: ***@***.***>

[ZhangGongjie]

感谢提问，抱歉回复的有些慢，最近太忙了 T T

以 MS-COCO 数据集为例，训练的指令是 ./scripts/run_experiments_coco.sh，其实Line9-24是base-training stage，Line48-67是few-shot-fintuning stage。Line48-67的命令结束后将自动执行few-shot inference。

如果要单独在novel classes上执行 inference ，直接可以执行：

python -u main.py \
--dataset_file coco_base \
--backbone resnet101 \
--num_feature_levels 1 \
--enc_layers 6 \
--dec_layers 6 \
--hidden_dim 256 \
--num_queries 300 \
--batch_size 2 \
--category_codes_cls_loss \
--resume path/to/checkpoint.pth/generated/by/few-shot-fintuning \
--fewshot_finetune \
--fewshot_seed ${fewshot_seed} \
--num_shots ${num_shot} \
--eval \
2>&1 | tee ${FS_FT_DIR}/log_inference.txt

其中加粗部分为需要需要特别注意的指令。希望可以帮到你！

[ZhangGongjie]

Question: After training, how to perform inference?

Answer:

Let's take MS-COCO as an example. The training & inference scripts are in ./scripts/run_experiments_coco.sh, of which Lines 9-24 refer to "base-training stage", and Lines 48-67 refer to "few-shot-finetuning stage". Lines 48-67 will also perform inference after finetuning is done.

If user wishes to perform inference, simply run:

python -u main.py \
--dataset_file coco_base \
--backbone resnet101 \
--num_feature_levels 1 \
--enc_layers 6 \
--dec_layers 6 \
--hidden_dim 256 \
--num_queries 300 \
--batch_size 2 \
--category_codes_cls_loss \
--resume path/to/checkpoint.pth/generated/by/few-shot-fintuning \
--fewshot_finetune \
--fewshot_seed ${fewshot_seed} \
--num_shots ${num_shot} \
--eval \
2>&1 | tee ${FS_FT_DIR}/log_inference.txt

Note that user should set --eval and --resume path/to/checkpoint.pth/generated/by/few-shot-fintuning correctly.