TODO.md

TODO

Day	Date	Tasks
1	Mon 29th Jul	Planned a detailed timeline by tasks per day. Download ART-Net Dataset. Convert ART-Net segmentation maps to bounding boxes for the entire tool and tooltip and prepare data in YOLO format. Begin training the YOLOv10 model on detection on the ART-Net dataset.l
2	Tues 30th Jul	Train a more bulky model on the ART-Net dataset. Adapt the ART-Net method as the anchor-free approach, validate reproducibility and then test on the ART-Net dataset.
3	Wed 31st Jul	Adapt the annotation software to annotate bounding box and tool tips.
4	Thurs 1st Aug	Transition the detection problem to tracking. We need a way to consider IDs for each tool and deal with disappearing tools. YOLO should already deal with this, but we would need to adapt the ART-Net method here.
5	Fri 2nd Aug	Partially labelled video 5.
6	Sat 3rd Aug	Partially labelled video 5.
7	Sun 4th Aug	Partially labelled video 5.
8	Mon 5th Aug	~~Fully labelled video 5. Ideally, out of 24 videos (107,698 images), we will have 23 semi-labelled with 1 image (1% labelled data) every 100 frames (1,048 total images), where we can employ a cross-validation split of ~80/20 (839/209) and a 2680-image test set (1 video exclusive for testing, completely labelled so that we can evaluate the model on unseen data and be confident in the result and see difference temporally).~~
9	Tues 6th Aug	Finish labelling the training set. Apply the YOLO model as the anchor-based approach to the in-house dataset. Also perform some CV techniques to help extract the tool, e.g. background removal, test with different losses and base architecture models (ResNet instead of VGG).
10	Wed 7th Aug	Ideally, development will be finished by this date, though models may still be training. Labelled tool tips in training data.
11	Thurs 8th Aug	Finished labelling tool tips in test data.
12	Fri 9th Aug	Train SIMO on ART-Net dataset and implement tracking with YOLO.
13	Sat 10th Aug	Run YOLO on test set and attempt to fix tracking issues. Begin writing the report with a draft abstract.
14	Sun 11th Aug	The SIMO (anchor-free model) needs to be trained and evaluated on the ART-Net test sets with key metrics extracted.
15	Mon 12th Aug	The SIMO (anchor-free model) needs to be trained and evaluated on the 6DOF test sets with key metrics extracted. Add outline document notes.
16	Tues 13th Aug	Fix YOLO tracking issue. Train new SIMO model on ART-Net, 6DOF and obtained tracking videos.
17	Wed 14th Aug	Generated all annotations using YOLOv10X. Create table of results. YOLOv8 models on 6DOF.
18	Thurs 15th Aug	Processed annotations for RetinaNet anchor-box optimisation. YOLOv8 models on ART-Net. Run RetinaNet (with and without anchor-box optimisation) on 6DOF and ART-Net.
19	Fri 16th Aug	Prepared models and data for running future experiments.
20	Sat 17th Aug	Run remaining RetinaNet models. Run EfficientDet models. Introduction.
21	Sun 18th Aug	Run remaining EfficientDet models and produce videos. Rerun validation on YOLOv8 and YOLOv10 on 6DOF and ART-Net at 0.5 IoU and get non-tracked videos. Improve tracking and get new tracking videos. Scheduled running SIMO models. Introduction.
22	Mon 19th Aug	Run DETR on ART. Computer vision and AI methods. PRISMA diagram. Added notes from Google Sheets.
23	Tues 20th Aug	Schematic diagram inspired by CholecTrack20. Add remaining notes from Google Sheets. Run DETR on 6DOF. Formulated PhD Plan. Results tables.
24	Wed 21st Aug	Fix repo. Presentation.
25	Thurs 22nd Aug	Introduction.
26	Fri 23rd Aug	Results and partial discussion.
27	Sat 24th Aug	Complete discussion and partial methodology.
28	Sun 25th Aug	Partial methodology.
29	Mon 26th Aug	Complete methodology.
30	Tues 27th Aug	Start background research.
31	Wed 28th Aug	Continue background research.
32	Thurs 29th Aug	Continue background research.
33	Fri 30th Aug	Finish background research. If not enough references, use "Selected Papers" in the Google Sheets. Report deadline.