search

YangRui2015 / RiC

Code for the ICML 2024 paper "Rewards-in-Context: Multi-objective Alignment of Foundation Models with Dynamic Preference Adjustment"
48 stars 4 forks source link

Missing file in ric_txt2img directory #13

Open EvaFlower opened 1 month ago

EvaFlower commented 1 month ago

Hi,

May you provide the file indicated in the train.sh export TRAIN_JSON="/mnt/aigc_cq/private/amandaaluo/dataset/multi_obj/multi_obj_with_clip_base.json" .

Also, I wonder if you could release your trained model checkpoints for ease of evaluation.

Best,

amandaluof commented 1 month ago

Hi,

May you provide the file indicated in the train.sh export TRAIN_JSON="/mnt/aigc_cq/private/amandaaluo/dataset/multi_obj/multi_obj_with_clip_base.json" .

Also, I wonder if you could release your trained model checkpoints for ease of evaluation.

Best,

Hi, we have uploaded our train_json at https://drive.google.com/file/d/1QH_nZWSbPlrpRdtTI-mxzqC7lK2HlON6/view?usp=sharing and for the checkpoint we may release it later.

EvaFlower commented 1 month ago

Hi, we have uploaded our train_json at https://drive.google.com/file/d/1QH_nZWSbPlrpRdtTI-mxzqC7lK2HlON6/view?usp=sharing and for the checkpoint we may release it later.

@amandaluof I really appreciate your follow-up.

I am still getting stuck in reproducing the results of the text2image part. In the multi_obj_with_clip_base.json, I can see you retrieve data from LAION-625K/improved_aesthetics_6.5plus/processed_data/images/00013_00862.jpg and LAION/laion_1.2b_subset/images/ec910b35-5b8c-4320-8c4f-7e6a20fe4d11.png.

Can you give some instructions about how you obtain 120k training image pairs from LAION-5B?

amandaluof commented 2 weeks ago

Thank you for the reminder. We have added instructions on how to prepare the training data here.