Segment Anything Model (#552)

* add segment anything model

* add readme

* reorg file structure

* update

* lint

* minor updates

* ack

* fix weight loading

* simplify

* fix to run notebooks

* amg in mlx

* remove torch dependency

* nit in README

* return indices in nms

* simplify

* bugfix / simplify

* fix bug'

* simplify

* fix notebook and remove output

* couple more nits

---------

Co-authored-by: Awni Hannun <awni@apple.com>

segment_anything/README.md

@@ -0,0 +1,39 @@
+# Segment Anything
+
+An implementation of the Segment Anything Model (SAM) in MLX. See the original
+repo by Meta AI for more details.[^1]
+
+## Installation
+
+```bash
+pip install -r requirements.txt
+```
+
+## Convert
+
+```bash
+python convert.py --hf-path facebook/sam-vit-base --mlx-path sam-vit-base
+```
+
+The `safetensors` weight file and configs are downloaded from Hugging Face,
+converted, and saved in the directory specified by `--mlx-path`.
+
+The model sizes are:
+
+- `facebook/sam-vit-base`
+- `facebook/sam-vit-large`
+- `facebook/sam-vit-huge`
+
+## Run
+
+See examples `notebooks/predictor_example.ipynb` and
+`notebooks/automatic_mask_generator_example.ipynb` to try the Segment Anything
+Model with MLX.
+
+You can also generate masks from the command line:
+
+```bash
+python main.py --model <path/to/model> --input <image_or_folder> --output <path/to/output>
+```
+
+[^1]: The original Segment Anything [GitHub repo](https://github.com/facebookresearch/segment-anything/tree/main).