ChengYue's Cat

待补实验一

• gama和walk_num超参数实验
  • 因为使用了超参数\(\gamma\)控制混合样本时的相似度权重，因此需要取gamma=0.0(mix unsim), 0.2, 0.5(random sim, same weight), 0.8, 1.0(mix sim). 当前对于CUB和AWA数据集还需要补gamma=0.2, 0.8， walk_num=2, 4, 6, 8, 10的实验。

    nohup python train_awa2.py --opts 
                                    gama 0.2 
                                    fine_iter_num 50 
                                    fine_epoch_num 500 
                                    walk_num 2 
                                    key_word label_fintune 
                                    temp 5.0 
                                    init_s2v 2 pkl_data /data/data/pkl_data_coarse_mix/V2S_CLS_AWA2_448_11_lr_0.0005_decay_0.001_data.pkl 
                                    >>awa2_label_finetune.txt 2>&1 &
    

    AWA--gama=0.2--walk_num=2 --2021-08-21 20:20:40,627 - [Fine|210]:::ZSL-S: 96.01:::ZSL-U: 77.25:::GZSL-S: 84.44:::GZSL-U: 71.48:::GZSL-H: 77.42:::Calibration_Factor:0.91:::Train Time:-2.04:::Test Time: 1.58 -/data/code/C2F_ZSL/log_coarse_final/AWA2/Fine/label_fintune/GMIX_0.2_lr_0.0005_decay_0.0001_temp_5.0_pkl_data_V2S_CLS_AWA2_448_11_lr_0.0005_decay_0.001_data.pkl_walk_num_2_2021-08-21-20-03-55.txt

    AWA--gama=0.2--walk_num=4 2021-08-21 20:43:12,897 - [Fine|55]:::ZSL-S: 95.58:::ZSL-U: 78.62:::GZSL-S: 83.22:::GZSL-U: 72.97:::GZSL-H: 77.76:::Calibration_Factor:0.89:::Train Time:-2.17:::Test Time: 1.49 /data/code/C2F_ZSL/log_coarse_final/AWA2/Fine/label_fintune/GMIX_0.2_lr_0.0005_decay_0.0001_temp_5.0_pkl_data_V2S_CLS_AWA2_448_11_lr_0.0005_decay_0.001_data.pkl_walk_num_4_2021-08-21-20-38-13.txt

    AWA--gama=0.2--walk_num=6 2021-08-21 21:09:48,376 - [Fine|47]:::ZSL-S: 95.84:::ZSL-U: 79.41:::GZSL-S: 83.80:::GZSL-U: 74.01:::GZSL-H: 78.60:::Calibration_Factor:0.89:::Train Time:-1.64:::Test Time: 1.23 2021-08-21 21:29:31,319 - current best::::::ZSL_U: 80.06:::::::::ZSL_H: 78.60::::::::ZSL_H_Epoch: 47 /data/code/C2F_ZSL/log_coarse_final/AWA2/Fine/label_fintune/GMIX_0.2_lr_0.0005_decay_0.0001_temp_5.0_pkl_data_V2S_CLS_AWA2_448_11_lr_0.0005_decay_0.001_data.pkl_walk_num_6_2021-08-21-21-05-59.txt

    AWA--gama=0.2--walk_num=8 2021-08-21 22:02:18,567 - [Fine|18]:::ZSL-S: 95.49:::ZSL-U: 77.64:::GZSL-S: 83.28:::GZSL-U: 70.95:::GZSL-H: 76.62:::Calibration_Factor:0.89:::Train Time:-2.19:::Test Time: 1.47 2021-08-21 22:22:04,870 - current best::::::ZSL_U: 78.57:::::::::ZSL_H: 76.62::::::::ZSL_H_Epoch: 18 /data/code/C2F_ZSL/log_coarse_final/AWA2/Fine/label_fintune/GMIX_0.2_lr_0.0005_decay_0.0001_temp_5.0_pkl_data_V2S_CLS_AWA2_448_11_lr_0.0005_decay_0.001_data.pkl_walk_num_8_2021-08-21-21-59-08.txt

    AWA--gama=0.2--walk_num=10 2021-08-21 22:32:44,675 - [Fine|36]:::ZSL-S: 95.54:::ZSL-U: 79.27:::GZSL-S: 85.23:::GZSL-U: 71.96:::GZSL-H: 78.04:::Calibration_Factor:0.91:::Train Time:-2.16:::Test Time: 1.49 2021-08-21 22:51:23,492 - current best::::::ZSL_U: 80.24:::::::::ZSL_H: 78.04::::::::ZSL_H_Epoch: 36 /data/code/C2F_ZSL/log_coarse_final/AWA2/Fine/label_fintune/GMIX_0.2_lr_0.0005_decay_0.0001_temp_5.0_pkl_data_V2S_CLS_AWA2_448_11_lr_0.0005_decay_0.001_data.pkl_walk_num_10_2021-08-21-22-28-20.txt

    AWA--gama=0.8--walk_num=2 2021-08-21 23:21:14,953 - current best::::::ZSL_U: 76.77:::::::::ZSL_H: 76.33::::::::ZSL_H_Epoch: 9 2021-08-21 23:00:38,205 - [Fine|9]:::ZSL-S: 95.46:::ZSL-U: 76.32:::GZSL-S: 82.92:::GZSL-U: 70.71:::GZSL-H: 76.33:::Calibration_Factor:0.87:::Train Time:-2.25:::Test Time: 1.76

    AWA--gama=0.8--walk_num=4 2021-08-21 23:37:25,098 - current best::::::ZSL_U: 77.86:::::::::ZSL_H: 77.75::::::::ZSL_H_Epoch: 279 2021-08-21 23:35:44,513 - [Fine|279]:::ZSL-S: 96.10:::ZSL-U: 77.48:::GZSL-S: 85.81:::GZSL-U: 71.07:::GZSL-H: 77.75:::Calibration_Factor:0.88:::Train Time:-1.94:::Test Time: 3.06

    AWA--gama=0.8--walk_num=6 2021-08-21 23:59:43,506 - current best::::::ZSL_U: 77.45:::::::::ZSL_H: 76.31::::::::ZSL_H_Epoch: 268 2021-08-21 23:57:06,875 - [Fine|268]:::ZSL-S: 96.24:::ZSL-U: 75.19:::GZSL-S: 84.01:::GZSL-U: 69.90:::GZSL-H: 76.31:::Calibration_Factor:0.88:::Train Time:-1.93:::Test Time: 3.05

    AWA--gama=0.8--walk_num=8 2021-08-21 23:59:42,924 - current best::::::ZSL_U: 78.23:::::::::ZSL_H: 77.28::::::::ZSL_H_Epoch: 210 2021-08-21 23:52:14,376 - [Fine|210]:::ZSL-S: 95.98:::ZSL-U: 76.32:::GZSL-S: 85.52:::GZSL-U: 70.50:::GZSL-H: 77.28:::Calibration_Factor:0.88:::Train Time:-1.91:::Test Time: 3.05

    AWA--gama=0.8--walk_num=10 2021-08-22 00:04:31,389 - current best::::::ZSL_U: 78.13:::::::::ZSL_H: 77.33::::::::ZSL_H_Epoch: 288 2021-08-22 00:04:05,826 - [Fine|288]:::ZSL-S: 95.93:::ZSL-U: 77.83:::GZSL-S: 84.38:::GZSL-U: 71.37:::GZSL-H: 77.33:::Calibration_Factor:0.86:::Train Time:-1.49:::Test Time: 0.92

  ---

  ---

  CUB--gama=0.2--walk_num=2,4,6,8,10

  2021-08-22 05:02:24,518 - current best::::::ZSL_U: 83.35:::::::::GZSL_H: 76.47::::::::GZSL_H_Epoch: 146 [Fine|146]:::ZSL-S: 86.55:::ZSL-U: 82.86:::GZSL-S: 78.23:::GZSL-U: 74.79:::GZSL-H: 76.47:::Calibration_Factor:0.97

  2021-08-22 05:41:53,889 - current best::::::ZSL_U: 82.98:::::::::GZSL_H: 76.16::::::::GZSL_H_Epoch: 194 [Fine|194]:::ZSL-S: 86.53:::ZSL-U: 82.92:::GZSL-S: 77.65:::GZSL-U: 74.73:::GZSL-H: 76.16:::Calibration_Factor:0.96

  2021-08-22 05:40:47,700 - current best::::::ZSL_U: 82.42:::::::::GZSL_H: 75.83::::::::GZSL_H_Epoch: 103 [Fine|103]:::ZSL-S: 86.02:::ZSL-U: 82.42:::GZSL-S: 77.57:::GZSL-U: 74.18:::GZSL-H: 75.83:::Calibration_Factor:0.96

  2021-08-22 05:42:04,034 - current best::::::ZSL_U: 82.22:::::::::GZSL_H: 75.67::::::::GZSL_H_Epoch: 166 [Fine|166]:::ZSL-S: 86.72:::ZSL-U: 82.03:::GZSL-S: 77.68:::GZSL-U: 73.76:::GZSL-H: 75.67:::Calibration_Factor:0.96

  2021-08-22 05:47:59,228 - current best::::::ZSL_U: 82.31:::::::::GZSL_H: 76.42::::::::GZSL_H_Epoch: 179 [Fine|179]:::ZSL-S: 86.69:::ZSL-U: 82.11:::GZSL-S: 78.87:::GZSL-U: 74.11:::GZSL-H: 76.42:::Calibration_Factor:0.96

  CUB--gama=0.8--walk_num=2,4,6,8,10

  2021-08-22 05:49:25,384 - current best::::::ZSL_U: 83.27:::::::::GZSL_H: 77.01::::::::GZSL_H_Epoch: 169 [Fine|169]:::ZSL-S: 87.03:::ZSL-U: 82.77:::GZSL-S: 78.11:::GZSL-U: 75.95:::GZSL-H: 77.01:::Calibration_Factor:0.96

  2021-08-22 05:47:52,126 - current best::::::ZSL_U: 83.57:::::::::GZSL_H: 77.03::::::::GZSL_H_Epoch: 132 [Fine|132]:::ZSL-S: 87.10:::ZSL-U: 83.08:::GZSL-S: 79.74:::GZSL-U: 74.50:::GZSL-H: 77.03:::Calibration_Factor:0.96

  2021-08-22 05:48:29,115 - current best::::::ZSL_U: 83.05:::::::::GZSL_H: 76.66::::::::GZSL_H_Epoch: 158 [Fine|158]:::ZSL-S: 87.68:::ZSL-U: 82.97:::GZSL-S: 78.78:::GZSL-U: 74.66:::GZSL-H: 76.66:::Calibration_Factor:0.96

  2021-08-22 05:46:33,733 - current best::::::ZSL_U: 83.16:::::::::GZSL_H: 76.48::::::::GZSL_H_Epoch: 119 [Fine|119]:::ZSL-S: 86.82:::ZSL-U: 82.12:::GZSL-S: 79.53:::GZSL-U: 73.66:::GZSL-H: 76.48:::Calibration_Factor:0.96

  2021-08-22 05:47:59,228 - current best::::::ZSL_U: 82.31:::::::::GZSL_H: 76.42::::::::GZSL_H_Epoch: 179 [Fine|179]:::ZSL-S: 86.69:::ZSL-U: 82.11:::GZSL-S: 78.87:::GZSL-U: 74.11:::GZSL-H: 76.42:::Calibration_Factor:0.96

• 对于SUN和FLO数据集还需要补所有gama和walk_num参数组合下的实验，==在本地机器上跑==.
  • nohup python train_sun.py --opts gama 0.0 walk_num 2 temp 1.0 >>sun.log 2>&1 &
  • nohup python train_sun.py --opts gama 0.0 walk_num 2 temp 5.0 >>sun.log 2>&1 &
  • nohup python train_sun.py --opts gama 0.0 walk_num 2 temp 0.2 >>sun.log 2>&1 &
  • nohup python train_sun.py --opts gama 1.0 walk_num 2 temp 1.0 >>sun.log 2>&1 &
  • nohup python train_sun.py --opts gama 1.0 walk_num 2 temp 5.0 >>sun.log 2>&1 &

  • nohup python train_sun.py --opts gama 1.0 walk_num 2 temp 0.2 >>sun.log 2>&1 &

待补实验二

• sparse max温度超参数实验
  • AWA数据集需要补上温度参数为temp=0.2, 2的实验,(walk_num=4或者6, gama=1.0即混合不相似的特征)
  • CUB数据集需要补上温度参数为temp=0.2, 0.5的实验, (walk_num=4或者6, gama=0.0, 即混合相似的特征)
  • SUN和FLO数据集需要补上温度参数为temp=0.2, 0.5, 1.0, 2, 5的实验，(walk_num=4或者6)

待补实验三

• 普通finetune后得到的特征，在c2f_zsl模型下得到的结果(四个结果)
  • AWA数据集已经fintune，可以直接找个数据集跑实验
  • CUB, FLO, SUN数据集需要fintune

使用了正交化损失对模型带来的提升(待定)

• CUB和AWA上应该不用做，要做的话只用做SUN和FLO数据集的。

This site is open source. Improve this page.