Instructions to use thetechgeekko/latent-android-models with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- LiteRT
How to use thetechgeekko/latent-android-models with LiteRT:
# No code snippets available yet for this library. # To use this model, check the repository files and the library's documentation. # Want to help? PRs adding snippets are welcome at: # https://github.com/huggingface/huggingface.js
- Notebooks
- Google Colab
- Kaggle
Add LATENT U2Netp conversion and legacy MiDaS fallback
Browse filesAdds exact hashes and per-artifact licensing; stock Google models remain pinned to official versioned downloads.
- MIDAS_MIT.txt +21 -0
- README.md +54 -1
- midas_v2_1_small_256_fp32.tflite +3 -0
- u2netp_d0_320_fp16.tflite +3 -0
MIDAS_MIT.txt
ADDED
|
@@ -0,0 +1,21 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
MIT License
|
| 2 |
+
|
| 3 |
+
Copyright (c) 2019 Intel ISL (Intel Intelligent Systems Lab)
|
| 4 |
+
|
| 5 |
+
Permission is hereby granted, free of charge, to any person obtaining a copy
|
| 6 |
+
of this software and associated documentation files (the "Software"), to deal
|
| 7 |
+
in the Software without restriction, including without limitation the rights
|
| 8 |
+
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
| 9 |
+
copies of the Software, and to permit persons to whom the Software is
|
| 10 |
+
furnished to do so, subject to the following conditions:
|
| 11 |
+
|
| 12 |
+
The above copyright notice and this permission notice shall be included in all
|
| 13 |
+
copies or substantial portions of the Software.
|
| 14 |
+
|
| 15 |
+
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
| 16 |
+
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
| 17 |
+
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
| 18 |
+
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
| 19 |
+
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
| 20 |
+
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
|
| 21 |
+
SOFTWARE.
|
README.md
CHANGED
|
@@ -1,5 +1,6 @@
|
|
| 1 |
---
|
| 2 |
-
license:
|
|
|
|
| 3 |
library_name: tflite
|
| 4 |
pipeline_tag: depth-estimation
|
| 5 |
tags:
|
|
@@ -17,6 +18,9 @@ This repository hosts model artifacts generated specifically for the LATENT
|
|
| 17 |
Android camera app. It intentionally does not contain LATENT's film LUTs or
|
| 18 |
Qualcomm QNN context binaries.
|
| 19 |
|
|
|
|
|
|
|
|
|
|
| 20 |
## DA3 Small — LiteRT FP32, 672 × 896
|
| 21 |
|
| 22 |
`da3_small_672x896_fp32.tflite` is a LATENT-generated conversion of
|
|
@@ -59,3 +63,52 @@ The pinned community converter is
|
|
| 59 |
See `NOTICE.md` for the modification and attribution record and `LICENSE` for
|
| 60 |
the Apache License 2.0. This model card records engineering provenance; it is
|
| 61 |
not a production-quality claim or legal opinion.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
+
license: other
|
| 3 |
+
license_name: mixed-apache-2.0-mit
|
| 4 |
library_name: tflite
|
| 5 |
pipeline_tag: depth-estimation
|
| 6 |
tags:
|
|
|
|
| 18 |
Android camera app. It intentionally does not contain LATENT's film LUTs or
|
| 19 |
Qualcomm QNN context binaries.
|
| 20 |
|
| 21 |
+
Each artifact below has its own immutable byte count, SHA-256, origin, and
|
| 22 |
+
license. A repository-level license must not be inferred for a different file.
|
| 23 |
+
|
| 24 |
## DA3 Small — LiteRT FP32, 672 × 896
|
| 25 |
|
| 26 |
`da3_small_672x896_fp32.tflite` is a LATENT-generated conversion of
|
|
|
|
| 63 |
See `NOTICE.md` for the modification and attribution record and `LICENSE` for
|
| 64 |
the Apache License 2.0. This model card records engineering provenance; it is
|
| 65 |
not a production-quality claim or legal opinion.
|
| 66 |
+
|
| 67 |
+
## U2Netp d0 — LATENT TFLite FP16 conversion, 320 x 320
|
| 68 |
+
|
| 69 |
+
`u2netp_d0_320_fp16.tflite` is Clearstate's format conversion of the plain
|
| 70 |
+
U2Netp salient-object checkpoint. It is not the separately trained
|
| 71 |
+
`u2net_portrait` model.
|
| 72 |
+
|
| 73 |
+
| Property | Value |
|
| 74 |
+
| --- | --- |
|
| 75 |
+
| Size | 2,378,688 bytes |
|
| 76 |
+
| SHA-256 | `357bd5214d6725efd88dc1578de5308f8a80c6df63ac3f35c2021e955f4a4bc5` |
|
| 77 |
+
| Precision | FP16 weights, float32 I/O |
|
| 78 |
+
| Input | float32 NHWC `[1, 320, 320, 3]` |
|
| 79 |
+
| Output | d0 saliency, float32 NHWC `[1, 320, 320, 1]` |
|
| 80 |
+
| Upstream architecture | `xuebinqin/U-2-Net` |
|
| 81 |
+
| Source checkpoint | plain `u2netp` obtained through `danielgatis/rembg` |
|
| 82 |
+
| Modification | pruned to d0 and converted ONNX to TFLite by Clearstate on 2026-07-19 |
|
| 83 |
+
| License | Apache-2.0; see `LICENSE` |
|
| 84 |
+
|
| 85 |
+
## MiDaS v2.1 Small — legacy Intel TFLite FP32
|
| 86 |
+
|
| 87 |
+
`midas_v2_1_small_256_fp32.tflite` is the unmodified legacy app fallback. It
|
| 88 |
+
is retained for exact clean-checkout reproduction while LATENT evaluates the
|
| 89 |
+
smaller revision-pinned LiteRT Community conversion.
|
| 90 |
+
|
| 91 |
+
| Property | Value |
|
| 92 |
+
| --- | --- |
|
| 93 |
+
| Size | 66,338,288 bytes |
|
| 94 |
+
| SHA-256 | `93d871071edff1218973ce25ee27ce95ccd20450c70a55e1b89efa3f5a772cdd` |
|
| 95 |
+
| Precision | FP32 |
|
| 96 |
+
| Input | float32 NHWC `[1, 256, 256, 3]` |
|
| 97 |
+
| Output | relative inverse depth `[1, 256, 256]` |
|
| 98 |
+
| Upstream | Intel MiDaS v2.1 Small TFLite |
|
| 99 |
+
| Modification | none by Clearstate |
|
| 100 |
+
| License | MIT; see `licenses/MIDAS_MIT.txt` |
|
| 101 |
+
|
| 102 |
+
## Stock Google MediaPipe models
|
| 103 |
+
|
| 104 |
+
LATENT does not mirror Google's stock model files here. Clean checkouts fetch
|
| 105 |
+
the exact versioned Google objects and verify these identities before staging:
|
| 106 |
+
|
| 107 |
+
| App artifact | Official versioned object | Size | SHA-256 | License evidence |
|
| 108 |
+
| --- | --- | ---: | --- | --- |
|
| 109 |
+
| `gesture_recognizer.task` | `gesture_recognizer/gesture_recognizer/float16/1` | 8,373,440 | `97952348cf6a6a4915c2ea1496b4b37ebabc50cbbf80571435643c455f2b0482` | Google MediaPipe model bundle; see the linked upstream model card in LATENT's lock file |
|
| 110 |
+
| `magic_touch.tflite` | `interactive_segmenter/magic_touch/float32/1` | 6,227,884 | `e24338a717c1b7ad8d159666677ef400babb7f33b8ad60c4d96db4ecf694cd25` | Apache-2.0 per Google's MagicTouch model card |
|
| 111 |
+
|
| 112 |
+
The Qualcomm-generated QNN context binaries are device-specific build inputs
|
| 113 |
+
and are stored separately in a private artifact repository. They are not
|
| 114 |
+
licensed or presented here as generally reusable models.
|
midas_v2_1_small_256_fp32.tflite
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:93d871071edff1218973ce25ee27ce95ccd20450c70a55e1b89efa3f5a772cdd
|
| 3 |
+
size 66338288
|
u2netp_d0_320_fp16.tflite
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:357bd5214d6725efd88dc1578de5308f8a80c6df63ac3f35c2021e955f4a4bc5
|
| 3 |
+
size 2378688
|