Addition of SAM2.1 Annotator for Instance Segmentation

[HonzaCuhel]

This PR

• adds the SAM2.1 image annotator for instance segmentation
• adds valid links for Luxonis logo in the .ipynb

SAM2.1 Info

Model	# Params (M)	Designation in DD
SAM2.1 Hiera Base Plus	80.8	base
SAM2.1 Hiera Large	224.4	large

SAM2.1 Latency

Here are the measured times per prompt for various batch sizes of SLM:

Model	Latency [ms]
SAM2.1 Hiera Base Plus	67.49±11.53
SAM2.1 Hiera Large	123.85±11.68
SlimSAM Base	206.01±3.20
SlimSAM Large	224.14±2.17

*Times are measured and averaged across 100 runs per 1 image for batch size of 4 on a JH device with 1 NVIDIA L4 GPU (24GB), 16 CPUs and 30 GB RAM.
** I measured the SlimSAM times several times, so the numbers should be correct, but it's definitely interesting. The SAM2 uses torch.bfloat16 during annotation + supports batch annotation, and performance looks better.

Examples

[github-actions]

Test Results

  6 files    6 suites   3h 16m 44s ⏱️
 77 tests  58 ✅  19 💤 0 ❌
462 runs  344 ✅ 118 💤 0 ❌

Results for commit f8ef89c.

♻️ This comment has been updated with latest results.

[sokovninn]

Have you benchmarked SlimSAM in a similar way? I wonder how they compare.

[HonzaCuhel]

Yeah, I did, I'll share the results later.

[github-actions]

☂️ Python Coverage

current status: ✅

Overall Coverage

Lines	Covered	Coverage	Threshold	Status
1852	1155	62%	0%	🟢

New Files

File	Coverage	Status
datadreamer/dataset_annotation/sam2_annotator.py	75%	🟢
TOTAL	75%	🟢

Modified Files

File	Coverage	Status
datadreamer/dataset_annotation/init.py	100%	🟢
datadreamer/pipelines/generate_dataset_from_scratch.py	38%	🟢
datadreamer/utils/config.py	100%	🟢
TOTAL	79%	🟢

updated for commit: f8ef89c by action🐍