Prebuilt 448x560 TensorRT engines: Jetson Orin (TRT 10.7) and RTX 50-series (TRT 10.16)
Browse files
.gitattributes
CHANGED
|
@@ -34,3 +34,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
|
| 34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
| 35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
| 36 |
da2_metric_hypersim_vits_364x448_trt100700_sm87.engine filter=lfs diff=lfs merge=lfs -text
|
|
|
|
|
|
|
|
|
| 34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
| 35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
| 36 |
da2_metric_hypersim_vits_364x448_trt100700_sm87.engine filter=lfs diff=lfs merge=lfs -text
|
| 37 |
+
da2_metric_hypersim_vits_448x560_trt100700_sm87.engine filter=lfs diff=lfs merge=lfs -text
|
| 38 |
+
da2_metric_hypersim_vits_448x560_trt101601_sm120.engine filter=lfs diff=lfs merge=lfs -text
|
README.md
CHANGED
|
@@ -16,8 +16,9 @@ Depth Anything V2 Small, metric, Hypersim (indoor, max depth 20 m), converted fr
|
|
| 16 |
(converted by `tools/convert_weights.py`; includes the unused ImageNet classifier head candle's dinov2 loads).
|
| 17 |
- `da2_metric_hypersim_vits.onnx`: an ONNX export taking any input size that is a multiple of 14, for TensorRT
|
| 18 |
(made by the crate's `tools/export_onnx.py`). `da2_metric_hypersim_vits_364x448.onnx` is the older fixed-size one.
|
| 19 |
-
-
|
| 20 |
-
Jetson AGX Orin
|
| 21 |
-
|
|
|
|
| 22 |
|
| 23 |
The crate pins these files by sha256 in its `model.json`. Weights are Apache-2.0, as the upstream Small model.
|
|
|
|
| 16 |
(converted by `tools/convert_weights.py`; includes the unused ImageNet classifier head candle's dinov2 loads).
|
| 17 |
- `da2_metric_hypersim_vits.onnx`: an ONNX export taking any input size that is a multiple of 14, for TensorRT
|
| 18 |
(made by the crate's `tools/export_onnx.py`). `da2_metric_hypersim_vits_364x448.onnx` is the older fixed-size one.
|
| 19 |
+
- Prebuilt fp16 TensorRT engines for a 448x560 input (dimos's default), so these GPUs skip the minutes-long engine build:
|
| 20 |
+
`..._448x560_trt100700_sm87.engine` (TensorRT 10.7, Jetson AGX Orin, JetPack 6 / L4T R36.3) and
|
| 21 |
+
`..._448x560_trt101601_sm120.engine` (TensorRT 10.16, RTX 50-series). Other TensorRT versions, GPUs or sizes reject
|
| 22 |
+
them and the crate builds its own from the ONNX. (`..._364x448_trt100700_sm87.engine` is an older fixed-size one.)
|
| 23 |
|
| 24 |
The crate pins these files by sha256 in its `model.json`. Weights are Apache-2.0, as the upstream Small model.
|
da2_metric_hypersim_vits_448x560_trt100700_sm87.engine
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:be974849b8704bc795254e248610a0408371cbdafdab29e2411953b9b0e2b4f9
|
| 3 |
+
size 55422620
|
da2_metric_hypersim_vits_448x560_trt101601_sm120.engine
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:7b9defa44791e09faac986b8163b6fd7c822b78a6f02dc24a4d6c11d541e9147
|
| 3 |
+
size 53640260
|