---
title: sapiens2-pointmap-0.4b-bf16
canonical_url: "https://www.modelscope.cn/models/mlx-community/sapiens2-pointmap-0.4b-bf16"
md_url: "https://www.modelscope.cn/models/mlx-community/sapiens2-pointmap-0.4b-bf16.md"
repository: mlx-community/sapiens2-pointmap-0.4b-bf16
last_updated: 2026-09-14
license: other
model_type:
  - sapiens2
architectures:
  - Sapiens2ForPointmapEstimation
base_model:
  - facebook/sapiens2-pointmap-0.4b
base_model_relation: finetune
parameters: 528.7M
tensor_type:
  - BF16
library_name:
  - mlx
  - safetensors
  - pytorch
frameworks:
  - pytorch
downloads: 4
stars: 0
tags:
  - sapiens
  - sapiens2
  - human-centric
  - pointmap
  - mlx
  - bf16
---

# sapiens2-pointmap-0.4b-bf16

> sapiens2-pointmap-0.4b-bf16 - mlx-community 在 ModelScope 开源的模型。mlx-community/sapiens2-pointmap-0.4b-bf16

mlx-community/sapiens2-pointmap-0.4b-bf16 是 ModelScope 魔搭社区上的 528.7M 参数机器学习模型，采用 other 许可，基于 facebook/sapiens2-pointmap-0.4b 构建。

- **Repository**: mlx-community/sapiens2-pointmap-0.4b-bf16
- **License**: other
- **Parameters**: 528.7M
- **Base model**: facebook/sapiens2-pointmap-0.4b
- **Tags**: sapiens, sapiens2, human-centric, pointmap, mlx, bf16
- **Downloads**: 4
- **Stars**: 0
- **Last updated**: 2026-09-14

Source: https://www.modelscope.cn/models/mlx-community/sapiens2-pointmap-0.4b-bf16

---

# mlx-community/sapiens2-pointmap-0.4b-bf16

bf16 MLX conversion of [`facebook/sapiens2-pointmap-0.4b`](https://huggingface.co/facebook/sapiens2-pointmap-0.4b) (Meta's
Sapiens2, ICLR 2026): per-pixel XYZ pointmaps + scale. Converted with mlx-vlm
0.7.0; the original float32 checkpoint is 2x this size.

What is in `model.safetensors` (1.06 GB):

- every parameter in bfloat16 (the reference runs inference in bf16 mixed
  precision);
- the q/k/v projections merged into one `wqkv` tensor per block, as the
  mlx-vlm Sapiens2 model expects.

Refer to the [original model card](https://huggingface.co/facebook/sapiens2-pointmap-0.4b) for the
model description, intended use and license.

## Use with mlx-vlm

```bash
pip install -U mlx-vlm
```

```python
from mlx_vlm import load
from mlx_vlm.models.sapiens2.generate import Sapiens2Predictor, read_image

model, _ = load("mlx-community/sapiens2-pointmap-0.4b-bf16")
predictor = Sapiens2Predictor(model)
output = predictor.infer(read_image("image.jpg"))
pointmaps, scale = output["pointmaps"], output["scales"]  # (H, W, 3), (1,)
```

Outputs are numpy arrays at the input resolution (dense tasks) or in
source-image pixel coordinates (pose). See the mlx-vlm Sapiens2 README for
preprocessing details and the per-task output keys.