Timesformer_keyframe_v1_wlasl100

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

• Loss: 2.2703
• Accuracy: 0.4556
• Precision: 0.4570
• Recall: 0.4556
• F1: 0.4259

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: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• training_steps: 3600
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	Precision	Recall	F1
No log	0.025	90	4.6040	0.0148	0.0013	0.0148	0.0024
4.6234	1.0249	180	4.5067	0.0207	0.0011	0.0207	0.0019
4.5373	2.0249	270	4.2366	0.0473	0.0265	0.0473	0.0287
4.1621	3.0251	361	3.7396	0.1923	0.1339	0.1923	0.1316
3.5688	4.025	451	3.4504	0.2278	0.2062	0.2278	0.1796
2.9066	5.0249	541	3.1612	0.3166	0.3083	0.3166	0.2673
2.2771	6.0249	631	2.9182	0.3373	0.3145	0.3373	0.2870
1.8115	7.0251	722	2.7938	0.3698	0.3761	0.3698	0.3347
1.4682	8.025	812	2.5983	0.3964	0.3813	0.3964	0.3569
1.051	9.0249	902	2.5988	0.3580	0.3465	0.3580	0.3162
1.051	10.0249	992	2.4744	0.4201	0.4105	0.4201	0.3828
0.8008	11.0251	1083	2.4348	0.4142	0.3949	0.4142	0.3748
0.618	12.025	1173	2.4154	0.4172	0.4113	0.4172	0.3778
0.4724	13.0249	1263	2.3397	0.4408	0.4542	0.4408	0.4125
0.3581	14.0249	1353	2.3164	0.4586	0.4777	0.4586	0.4280
0.2848	15.0251	1444	2.2820	0.4615	0.4822	0.4615	0.4387
0.2046	16.025	1534	2.3310	0.4379	0.4690	0.4379	0.4129
0.1561	17.0249	1624	2.3131	0.4556	0.4716	0.4556	0.4291
0.1236	18.0249	1714	2.2300	0.4527	0.4614	0.4527	0.4266
0.0903	19.0251	1805	2.3158	0.4438	0.4338	0.4438	0.4095
0.0903	20.025	1895	2.2703	0.4556	0.4570	0.4556	0.4259

Framework versions

• Transformers 4.44.0
• Pytorch 1.11.0+cu102
• Datasets 2.21.0
• Tokenizers 0.19.1