SAM 2 Update 12/11/2024 -- full model compilation for a major VOS speedup and a new SAM2VideoPredictor to better handle multi-object tracking (#486)

This PR provides new features and updates for SAM 2:

- We now support `torch.compile` of the entire SAM 2 model on videos, which can be turned on by setting `vos_optimized=True` in `build_sam2_video_predictor` (it uses the new `SAM2VideoPredictorVOS` predictor class in `sam2/sam2_video_predictor.py`).
  * Compared to the previous setting (which only compiles the image encoder backbone), the new full model compilation gives a major speedup in inference FPS.
  * In the VOS prediction script `tools/vos_inference.py`, you can specify this option in `tools/vos_inference.py` via the `--use_vos_optimized_video_predictor` flag.
  * Note that turning on this flag might introduce a small variance in the predictions due to numerical differences caused by `torch.compile` of the full model.
  * **PyTorch 2.5.1 is the minimum version for full support of this feature**. (Earlier PyTorch versions might run into compilation errors in some cases.) Therefore, we have updated the minimum PyTorch version to 2.5.1 accordingly in the installation scripts.
- We also update the implementation of the `SAM2VideoPredictor` class for the SAM 2 video prediction in `sam2/sam2_video_predictor.py`, which allows for independent per-object inference. Specifically, in the new `SAM2VideoPredictor`:
  * Now **we handle the inference of each object independently** (as if we are opening a separate session for each object) while sharing their backbone features.
  * This change allows us to relax the assumption of prompting for multi-object tracking. Previously (due to the batching behavior in inference), if a video frame receives clicks for only a subset of objects, the rest of the (non-prompted) objects are assumed to be non-existent in this frame (i.e., in such frames, the user is telling SAM 2 that the rest of the objects don't appear). Now, if a frame receives clicks for only a subset of objects, we do not make any assumptions about the remaining (non-prompted) objects (i.e., now each object is handled independently and is not affected by how other objects are prompted). As a result, **we allow adding new objects after tracking starts** after this change (which was previously a restriction on usage).
  * We believe that the new version is a more natural inference behavior and therefore switched to it as the default behavior. The previous implementation of `SAM2VideoPredictor` is backed up to in `sam2/sam2_video_predictor_legacy.py`. All the VOS inference results using `tools/vos_inference.py` should remain the same after this change to the `SAM2VideoPredictor` class.

INSTALL.md

@@ -2,7 +2,7 @@
 
 ### Requirements
 
-- Linux with Python ≥ 3.10, PyTorch ≥ 2.3.1 and [torchvision](https://github.com/pytorch/vision/) that matches the PyTorch installation. Install them together at https://pytorch.org to ensure this.
+- Linux with Python ≥ 3.10, PyTorch ≥ 2.5.1 and [torchvision](https://github.com/pytorch/vision/) that matches the PyTorch installation. Install them together at https://pytorch.org to ensure this.
   * Note older versions of Python or PyTorch may also work. However, the versions above are strongly recommended to provide all features such as `torch.compile`.
 - [CUDA toolkits](https://developer.nvidia.com/cuda-toolkit-archive) that match the CUDA version for your PyTorch installation. This should typically be CUDA 12.1 if you follow the default installation command.
 - If you are installing on Windows, it's strongly recommended to use [Windows Subsystem for Linux (WSL)](https://learn.microsoft.com/en-us/windows/wsl/install) with Ubuntu.
@@ -121,9 +121,9 @@ I got `undefined symbol: _ZN3c1015SmallVectorBaseIjE8grow_podEPKvmm` (or similar
 
 This usually happens because you have multiple versions of dependencies (PyTorch or CUDA) in your environment. During installation, the SAM 2 library is compiled against one version library while at run time it links against another version. This might be due to that you have different versions of PyTorch or CUDA installed separately via `pip` or `conda`. You may delete one of the duplicates to only keep a single PyTorch and CUDA version.
 
-In particular, if you have a lower PyTorch version than 2.3.1, it's recommended to upgrade to PyTorch 2.3.1 or higher first. Otherwise, the installation script will try to upgrade to the latest PyTorch using `pip`, which could sometimes lead to duplicated PyTorch installation if you have previously installed another PyTorch version using `conda`.
+In particular, if you have a lower PyTorch version than 2.5.1, it's recommended to upgrade to PyTorch 2.5.1 or higher first. Otherwise, the installation script will try to upgrade to the latest PyTorch using `pip`, which could sometimes lead to duplicated PyTorch installation if you have previously installed another PyTorch version using `conda`.
 
-We have been building SAM 2 against PyTorch 2.3.1 internally. However, a few user comments (e.g. https://github.com/facebookresearch/sam2/issues/22, https://github.com/facebookresearch/sam2/issues/14) suggested that downgrading to PyTorch 2.1.0 might resolve this problem. In case the error persists, you may try changing the restriction from `torch>=2.3.1` to `torch>=2.1.0` in both [`pyproject.toml`](pyproject.toml) and [`setup.py`](setup.py) to allow PyTorch 2.1.0.
+We have been building SAM 2 against PyTorch 2.5.1 internally. However, a few user comments (e.g. https://github.com/facebookresearch/sam2/issues/22, https://github.com/facebookresearch/sam2/issues/14) suggested that downgrading to PyTorch 2.1.0 might resolve this problem. In case the error persists, you may try changing the restriction from `torch>=2.5.1` to `torch==2.1.0` in both [`pyproject.toml`](pyproject.toml) and [`setup.py`](setup.py) to allow PyTorch 2.1.0.
 </details>
 
 <details>