jialicheng/ucf101_videomae-base

ucf101_42

This model is a fine-tuned version of MCG-NJU/videomae-base on the ucf101 dataset. It achieves the following results on the evaluation set:

• Loss: 0.8198
• Accuracy: 0.8298

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 8
• eval_batch_size: 64
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 20

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
No log	1.01	300	2.3345	0.4887
No log	2.01	600	1.4956	0.6433
No log	3.01	900	1.1647	0.7162
1.5784	4.01	1200	1.1531	0.7029
1.5784	5.01	1500	1.0251	0.7334
1.5784	6.01	1800	1.0315	0.7333
0.1821	7.01	2100	0.9787	0.7617
0.1821	8.01	2400	0.8933	0.7838
0.1821	9.01	2700	0.8781	0.7917
0.0651	10.01	3000	0.9051	0.7910
0.0651	11.01	3300	0.9593	0.7900
0.0651	12.01	3600	0.8054	0.8187
0.0651	13.01	3900	0.8679	0.8142
0.0265	14.01	4200	0.8380	0.8208
0.0265	15.01	4500	0.8317	0.8247
0.0265	16.01	4800	0.8027	0.8249
0.0091	17.01	5100	0.8240	0.8255
0.0091	18.01	5400	0.8480	0.8211
0.0091	19.01	5700	0.8198	0.8298
0.0091	19.87	5960	0.8315	0.8283

Framework versions

• Transformers 4.39.3
• Pytorch 2.2.2+cu118
• Datasets 2.18.0
• Tokenizers 0.15.2