SAM 2.1 Hiera-Large — INT8 Quantized

Meta's Segment Anything Model 2.1 (Hiera-Large backbone) quantized to INT8 for real-time robotic segmentation. 1.7x smaller — from 1.7 GB to 1.0 GB — with both image and video segmentation capabilities preserved.

This model is part of the RobotFlowLabs model library, built for the ANIMA agentic robotics platform — a modular ROS2-native AI system that brings foundation model intelligence to real robots operating in the real world.

Why This Model Exists

Robotic manipulation and navigation require pixel-precise understanding of the scene. SAM2 is the state-of-the-art for promptable segmentation — given a point, box, or mask prompt, it segments any object in images or tracks it through video. But at 1.7 GB, deploying SAM2 alongside other perception models on edge hardware eats precious VRAM.

We quantized SAM2.1 to INT8 and exported weights in SafeTensors format so robots can run segmentation in real-time alongside depth estimation, feature extraction, and action generation — all on a single edge GPU.

Model Details

Property	Value
Architecture	Hiera-Large vision backbone + SAM2 decoder
Input Resolution	1024 × 1024
Capabilities	Image segmentation, video object tracking
Mask Decoder	256-dim hidden, 8 attention heads, 3 multi-mask outputs
Memory Attention	4 layers, 2048-dim FFN, RoPE positional encoding
Memory Bank	7 frames temporal context
Original Model	`facebook/sam2.1-hiera-large`
License	Apache-2.0

Compression Results

Quantized on an NVIDIA L4 24GB GPU using INT8 dynamic quantization with SafeTensors export.

Metric	Original	INT8 Quantized	Change
Total Size	1,713 MB	1,038 MB	1.7x smaller
INT8 Weights	—	211 MB	Quantized linear layers
SafeTensors	—	828 MB	Full model weights
Quantization	FP32	INT8 Dynamic	Per-tensor symmetric
Format	PyTorch	SafeTensors + INT8 .pt	Dual format

Why SafeTensors instead of ONNX? SAM2 uses custom CUDA operations (roi_align, deformable attention) that aren't supported by the ONNX standard. SafeTensors provides fast, safe loading directly into PyTorch with zero-copy memory mapping.

Included Files

sam2.1-hiera-large-int8/
├── model_int8.pt              # 211 MB — INT8 quantized state dict
├── model.safetensors          # 828 MB — Full model in SafeTensors format
├── config.json                # Model configuration
├── preprocessor_config.json   # Image preprocessing config
└── README.md                  # This file

Quick Start

PyTorch (SafeTensors)

from transformers import Sam2Model, Sam2Processor
import torch

# Load with SafeTensors (automatic)
model = Sam2Model.from_pretrained("robotflowlabs/sam2.1-hiera-large-int8")
processor = Sam2Processor.from_pretrained("facebook/sam2.1-hiera-large")

model.to("cuda").eval()

# Segment with point prompt
inputs = processor(
    images=image,
    input_points=[[[500, 375]]],  # (x, y) point prompt
    return_tensors="pt"
).to("cuda")

with torch.no_grad():
    outputs = model(**inputs)

masks = processor.post_process_masks(
    outputs.pred_masks,
    inputs["original_sizes"],
    inputs["reshaped_input_sizes"]
)

INT8 Weights (Maximum Compression)

import torch
from transformers import Sam2Model

# Load architecture, then apply INT8 weights
model = Sam2Model.from_pretrained("facebook/sam2.1-hiera-large")
int8_state = torch.load("model_int8.pt", map_location="cuda", weights_only=True)
model.load_state_dict(int8_state, strict=False)

With FORGE (ANIMA Integration)

from forge.vision import VisionEncoderRegistry

# FORGE handles optimal loading and batching
segmenter = VisionEncoderRegistry.load("sam2.1-hiera-large-int8")
masks = segmenter.segment(image, points=[[500, 375]])

Use Cases in ANIMA

SAM2 is the segmentation backbone across multiple ANIMA modules:

Object Isolation — Segment graspable objects from cluttered scenes for manipulation planning
Workspace Mapping — Identify free space, obstacles, and surfaces for navigation
Video Tracking — Track objects across frames during manipulation sequences (7-frame temporal memory)
Safety Zones — Segment human body parts and keep-out regions for safe human-robot collaboration
Instance Separation — Distinguish individual objects when multiple similar items are present
Bin Picking — Segment individual parts from a bin for industrial pick-and-place

SAM2 Model Family

We provide all three SAM2.1 variants, optimized for different deployment scenarios:

Model	Params	Size	Speed	Best For
sam2.1-hiera-large-int8	Large	1.0 GB	Highest quality	Research, high-accuracy tasks
sam2.1-hiera-small-int8	Small	186 MB	Balanced	Production robotics
sam2.1-hiera-tiny-int8	Tiny	152 MB	Fastest	Real-time edge, Jetson Nano

About ANIMA

ANIMA is a modular, ROS2-native agentic robotics platform developed by RobotFlowLabs. It combines 58 specialized AI modules — from perception and planning to manipulation and safety — into a unified system that enables robots to understand, reason, and act in unstructured real-world environments.

Every foundation model in ANIMA must run on edge hardware (Jetson Orin, industrial PCs) under real-time constraints. That's why we built FORGE — our compression and distillation pipeline — and why we're releasing optimized model variants publicly.

We believe the robotics community deserves production-ready models, not just research checkpoints.

Other Collections

ANIMA Vision — SAM2, DINOv2, CLIP, SigLIP, Depth Anything
ANIMA Language — Qwen2.5, SmolLM2
ANIMA VLM — Qwen2.5-VL
ANIMA VLA — SmolVLA, RDT2-FM, FORGE students

Intended Use

Designed For

Promptable segmentation in robotic manipulation pipelines
Video object tracking during multi-step tasks
Instance segmentation for bin picking and object isolation
Real-time scene parsing on edge GPUs (Jetson Orin, L4)

Limitations

INT8 quantization may slightly reduce mask boundary precision on very fine structures
Video tracking requires sequential frame processing (not parallelizable)
Requires a prompt (point, box, or mask) — not a panoptic segmenter
Inherits biases from SA-V dataset (primarily indoor/outdoor natural scenes)

Out of Scope

Medical image segmentation without domain-specific validation
Autonomous driving perception (not trained on driving data)
Surveillance or tracking of individuals

Technical Details

Compression Pipeline

Original SAM2.1 Hiera-Large (FP32, 1.7 GB)
    │
    ├─→ torchao INT8 dynamic quantization (GPU-native)
    │   └─→ model_int8.pt (211 MB)
    │
    └─→ SafeTensors export (roi_align not ONNX-compatible)
        └─→ model.safetensors (828 MB)

Quantization: INT8 dynamic activation + INT8 weight via torchao on NVIDIA L4 GPU
Export: SafeTensors format — zero-copy memory mapping, fast loading, framework-agnostic
Why not ONNX: SAM2's roi_align and deformable attention are custom CUDA ops that ONNX opset 18 cannot represent
Hardware: NVIDIA L4 24GB, CUDA 13.0, PyTorch 2.10, Python 3.14

Attribution

Original Model: facebook/sam2.1-hiera-large by Meta AI (FAIR)
License: Apache-2.0 — free for commercial and research use
Paper: SAM 2: Segment Anything in Images and Videos — Ravi et al., 2024
Dataset: SA-V — 50.9K videos, 642.6K masklets
Compressed by: RobotFlowLabs using FORGE

Citation

@article{ravi2024sam2,
  title={SAM 2: Segment Anything in Images and Videos},
  author={Ravi, Nikhila and Gabeur, Valentin and Hu, Yuan-Ting and Hu, Ronghang and Ryali, Chaitanya and Ma, Tengyu and Khedr, Haitham and R{\"a}dle, Roman and Rolber, Chloe and Gustafson, Laura and others},
  journal={arXiv preprint arXiv:2408.00714},
  year={2024}
}

@misc{robotflowlabs2026anima,
  title={ANIMA: Agentic Networked Intelligence for Modular Autonomy},
  author={RobotFlowLabs},
  year={2026},
  url={https://huggingface.co/robotflowlabs}
}

Built with FORGE by RobotFlowLabs
Optimizing foundation models for real robots.