goktug14/yolov5_kvasir_polyp

yolov5m_kvasir_polyp

This model is a fine-tuned version of ultralytics/yolov5m on COCO dataset. It achieves the following results on the evaluation set in last epoch:

• Recall: 0.88479
• Precision: 0.9072
• mAP@50: 0.9328
• mAP@50-95: 0.7778
• Box Loss: 0.6825
• Cls Loss: 0.5279
• Dfl Loss: 1.1337

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

Kvasir dataset is an open-access collection focused on gastrointestinal polyps. It includes 1000 images, each accompanied by corresponding segmentation masks and bounding box annotations. The image resolutions within the dataset vary, ranging from 332x487 to 1920x1072 pixels. For our purposes, we randomly partitioned the images and their associated segmentation masks into training and testing sets, using an 80% allocation for training and 20% for testing.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.01
• learning_rate_factor: 0.01
• Optimizer: SGD
• train_batch_size: 16
• eval_batch_size: 16
• seed: 0
• lr_scheduler_type: linear
• num_epochs: 50

Training results

Epoch	Train Box Loss	Train Cls Loss	Train DFL Loss	Precision	Recall	mAP@50	mAP@50-95	Valid Box Loss	Valid Cls Loss	Valid DFL Loss
1	1.17657	2.36345	1.55172	0.39236	0.51152	0.42826	0.23057	1.42345	5.57235	2.02122
2	1.38219	1.78941	1.6711	0.08378	0.05991	0.02744	0.01038	3.19306	5.65541	3.36131
3	1.4393	1.80771	1.71444	0.17791	0.26728	0.11554	0.04333	2.31542	22.191	3.33876
4	1.46508	1.77948	1.72477	0.30202	0.25806	0.20997	0.10152	1.94347	5.65863	2.24456
5	1.41696	1.65393	1.68869	0.44164	0.20737	0.21563	0.12201	2.44439	2.57283	2.70957
6	1.34625	1.63672	1.64062	0.6144	0.64516	0.64903	0.40923	1.13771	3.31983	1.54363
7	1.31722	1.49886	1.62423	0.43942	0.34562	0.34845	0.15524	1.85409	3.20515	2.2562
8	1.27748	1.5087	1.58922	0.72454	0.65899	0.71588	0.4329	1.19252	1.82616	1.53101
9	1.22438	1.42816	1.55159	0.68385	0.63797	0.69648	0.43722	1.18662	1.50596	1.52833
10	1.21536	1.43968	1.5354	0.64495	0.59908	0.63943	0.43225	1.15523	2.03383	1.49961
...	...	...	...	...	...	...	...	...	...	...
41	0.73798	0.72353	1.20674	0.89355	0.84793	0.91711	0.73377	0.74772	0.68238	1.18388
42	0.70322	0.6215	1.17143	0.92202	0.84793	0.93047	0.74775	0.75578	0.61033	1.16188
43	0.69534	0.60391	1.16855	0.89488	0.86636	0.91829	0.73852	0.73135	0.62313	1.16575
44	0.72092	0.60997	1.19525	0.95241	0.83871	0.93116	0.75469	0.71709	0.58623	1.15658
45	0.67487	0.58592	1.15733	0.91481	0.84127	0.92245	0.74524	0.72996	0.5617	1.19229
46	0.70238	0.60048	1.19095	0.9224	0.87637	0.93562	0.75929	0.71903	0.54975	1.16705
47	0.63896	0.53985	1.14029	0.90244	0.90323	0.9365	0.76503	0.73252	0.54194	1.16863
48	0.63148	0.53112	1.1346	0.92686	0.88479	0.93843	0.76868	0.69789	0.53396	1.15055
49	0.61857	0.4978	1.11717	0.92212	0.87305	0.93688	0.77648	0.68321	0.53224	1.13785
50	0.59229	0.49861	1.08749	0.90728	0.88479	0.93281	0.77785	0.68251	0.52791	1.13379

Framework versions

• Numpy 2.0.2
• PIL 11.2.1
• Sklearn 1.6.1
• Ultralytics 8.3.120
• PyTorch 2.6.0+cu124