Surgical_instruments

Repo for images, annotations, notebooks and models for detecting surgical instruments.

Objective:

Develop an application that can reliably locate and identify surgical instruments in an image.

Workflow:

Surg_object_det_2021_10_14.pdf

Summary of Results:

A series of experiments including both supervised and unsupervised learning techniques showed that fine-tuning a VFNet model with a ResNet50_fpn_mstrain_2x backbone resulted in the best level of detection, with a mAP of 82.6, validation set loss of 0.94 and good performance on the held-out test set.

Self-training using teacher- and student-models did not perform as well as conventional supervised learning.

Deployment:

https://huggingface.co/spaces/yrodriguezmd/Surgical_instruments_app

Reports:

https://wandb.ai/maria_rodriguez/Transfer_learning_vf/reports/Transfer-Learning-Serial-versus-One-time-Training--VmlldzoxMjczNDMz

https://wandb.ai/maria_rodriguez/Surgical_instruments_models_/reports/Choosing-a-Model-for-Detecting-Surgical-Instruments--VmlldzoxMjI4NjQ0

Future Work:

Further improvement is needed in the detection of overlapping instruments/ dense distribution, as well as small objects (surgical needle).