Created by: velconia
object_detection 在本地进行过自测, 测试环境如下:
GPU: 4 * P40
数据集: pascalvoc
train.py 正常执行
infer.py 正常执行收敛情况如下:
train on pascalvoc with 16551 images
test on pascalvoc with 4952 images
Pass 0, batch 0, loss 35.34485626220703, time 4.745636940002441
Pass 0, batch 20, loss 11.94227123260498, time 5.786929130554199
Pass 0, batch 40, loss 8.538806915283203, time 5.208272933959961
Pass 0, batch 60, loss 7.659024238586426, time 5.300149440765381
Pass 0, batch 80, loss 7.16782283782959, time 4.942589044570923
Pass 0, batch 100, loss 6.835295677185059, time 6.101689338684082
Pass 0, batch 120, loss 6.667757034301758, time 5.5905067920684814
Pass 0, batch 140, loss 6.7157301902771, time 6.157291650772095
Pass 0, batch 160, loss 6.226722717285156, time 4.640245199203491
Pass 0, batch 180, loss 6.123575210571289, time 5.627285480499268
Pass 0, batch 200, loss 6.122291088104248, time 5.931992769241333
Pass 0, batch 220, loss 8.052138328552246, time 5.563196182250977
There are too few data to train on all devices.
Batch 0, map [0.41446137]
Batch 20, map [0.25457525]
Batch 40, map [0.24238662]
Batch 60, map [0.24026608]
save models to model/best_model
Pass 0, test map [0.23838544]
save models to model/0
Pass 1, batch 0, loss 5.957313537597656, time 5.037740230560303
Pass 1, batch 20, loss 5.859901428222656, time 5.262516498565674
Pass 1, batch 40, loss 5.776670455932617, time 5.448286533355713
Pass 1, batch 60, loss 6.2576398849487305, time 5.78664231300354
Pass 1, batch 80, loss 5.526543617248535, time 4.933773994445801
Pass 1, batch 100, loss 5.5869669914245605, time 5.409125089645386
Pass 1, batch 120, loss 5.736286163330078, time 6.599635124206543
Pass 1, batch 140, loss 6.025790214538574, time 6.25817084312439
Pass 1, batch 160, loss 5.804162979125977, time 5.377420902252197
Pass 1, batch 180, loss 5.4708251953125, time 5.602002382278442
Pass 1, batch 200, loss 5.320291519165039, time 6.41616678237915
Pass 1, batch 220, loss 5.235589027404785, time 6.089456796646118
Batch 0, map [0.5214782]
Batch 20, map [0.3836475]
Batch 40, map [0.38109452]
Batch 60, map [0.3807569]
save models to model/best_model本地自测text_classification (bow model) 模型收敛情况如下:
pass_id: 0, avg_acc: 0.849280, avg_cost: 0.352914
pass_id: 1, avg_acc: 0.915440, avg_cost: 0.216342
pass_id: 2, avg_acc: 0.929680, avg_cost: 0.182771
pass_id: 3, avg_acc: 0.939960, avg_cost: 0.162024
pass_id: 4, avg_acc: 0.947080, avg_cost: 0.147124
pass_id: 5, avg_acc: 0.952160, avg_cost: 0.134662
pass_id: 6, avg_acc: 0.957040, avg_cost: 0.125078
pass_id: 7, avg_acc: 0.960200, avg_cost: 0.116064
pass_id: 8, avg_acc: 0.964720, avg_cost: 0.108208
pass_id: 9, avg_acc: 0.967400, avg_cost: 0.100924
pass_id: 10, avg_acc: 0.971320, avg_cost: 0.093750
pass_id: 11, avg_acc: 0.973200, avg_cost: 0.088134
pass_id: 12, avg_acc: 0.975640, avg_cost: 0.082715
pass_id: 13, avg_acc: 0.978040, avg_cost: 0.077402
pass_id: 14, avg_acc: 0.980800, avg_cost: 0.071769
pass_id: 15, avg_acc: 0.983120, avg_cost: 0.067344
pass_id: 16, avg_acc: 0.984760, avg_cost: 0.062103
pass_id: 17, avg_acc: 0.985800, avg_cost: 0.058509
pass_id: 18, avg_acc: 0.987400, avg_cost: 0.054533
pass_id: 19, avg_acc: 0.989080, avg_cost: 0.050789
pass_id: 20, avg_acc: 0.990000, avg_cost: 0.047416
pass_id: 21, avg_acc: 0.991240, avg_cost: 0.043980
pass_id: 22, avg_acc: 0.992080, avg_cost: 0.040619
pass_id: 23, avg_acc: 0.992800, avg_cost: 0.037422
pass_id: 24, avg_acc: 0.994400, avg_cost: 0.034529
pass_id: 25, avg_acc: 0.994880, avg_cost: 0.032154
pass_id: 26, avg_acc: 0.995280, avg_cost: 0.029704
pass_id: 27, avg_acc: 0.996200, avg_cost: 0.027236
pass_id: 28, avg_acc: 0.996520, avg_cost: 0.025075
pass_id: 29, avg_acc: 0.997160, avg_cost: 0.023125infer结果如下:
model_path: bow_model/epoch0, avg_acc: 0.881120
model_path: bow_model/epoch1, avg_acc: 0.881800
model_path: bow_model/epoch2, avg_acc: 0.881600
model_path: bow_model/epoch3, avg_acc: 0.879440
model_path: bow_model/epoch4, avg_acc: 0.875120
model_path: bow_model/epoch5, avg_acc: 0.860640
model_path: bow_model/epoch6, avg_acc: 0.865920
model_path: bow_model/epoch7, avg_acc: 0.866840
model_path: bow_model/epoch8, avg_acc: 0.860680
model_path: bow_model/epoch9, avg_acc: 0.863480
model_path: bow_model/epoch10, avg_acc: 0.862000
model_path: bow_model/epoch11, avg_acc: 0.860040
model_path: bow_model/epoch12, avg_acc: 0.858200
model_path: bow_model/epoch13, avg_acc: 0.855640
model_path: bow_model/epoch14, avg_acc: 0.855840
model_path: bow_model/epoch15, avg_acc: 0.854440
model_path: bow_model/epoch16, avg_acc: 0.851560
model_path: bow_model/epoch17, avg_acc: 0.852720
model_path: bow_model/epoch18, avg_acc: 0.851280
model_path: bow_model/epoch19, avg_acc: 0.851440
model_path: bow_model/epoch20, avg_acc: 0.850840
model_path: bow_model/epoch21, avg_acc: 0.849720
model_path: bow_model/epoch22, avg_acc: 0.848520
model_path: bow_model/epoch23, avg_acc: 0.848520
model_path: bow_model/epoch24, avg_acc: 0.847600
model_path: bow_model/epoch25, avg_acc: 0.846000
model_path: bow_model/epoch26, avg_acc: 0.843120
model_path: bow_model/epoch27, avg_acc: 0.845720
model_path: bow_model/epoch28, avg_acc: 0.844760
model_path: bow_model/epoch29, avg_acc: 0.845520