search

lihong2303 / AGM

[ICCV2023] The repo for "Boosting Multi-modal Model Performance with Adaptive Gradient Modulation".
MIT License
24 stars 4 forks source link

Experiment parameters of av-mnist #12

Closed XKingham closed 2 weeks ago

XKingham commented 2 weeks ago

the result seems greatly different from the paper. How can i reproduce the result of av-mnist? Here is my input: python main.py --data_root '/home/tankedge/newdisk/xjh/avmnist' --device cuda:0 --methods AGM --modality Multimodal --fusion_type early_fusion --random_seed 999 --expt_dir checkpoint --expt_name test --batch_size 2048 --EPOCHS 100 --learning_rate 0.00005 --dataset AV-MNIST --alpha 2.5 1730442983328

XKingham commented 2 weeks ago

The train acc reached 0.69, however, the acc_a and acc_v still remained low, that's absurd