Publish validated merged OneWorld inference release
Browse filesPromote model bytes validated at 0866f9483f8199ee3026b6c5376098b4b12ade17, paired with GitHub ebd6f316ce8fe1da74650a9a5d5738e2ec85994a. Preserve all parameter values and precision; replace legacy rank shards with one named model.safetensors and link the published inference instructions.
- README.md +4 -15
- checkpoint/contract.json +7 -2
- checkpoint/rank00001.pt +0 -3
- checkpoint/rank00002.pt +0 -3
- checkpoint/rank00003.pt +0 -3
- checkpoint/rank00004.pt +0 -3
- checkpoint/rank00005.pt +0 -3
- checkpoint/rank00006.pt +0 -3
- checkpoint/rank00007.pt +0 -3
- checkpoint/rank00008.pt +0 -3
- checkpoint/rank00009.pt +0 -3
- checkpoint/rank00010.pt +0 -3
- checkpoint/rank00011.pt +0 -3
- checkpoint/rank00012.pt +0 -3
- checkpoint/rank00013.pt +0 -3
- checkpoint/rank00014.pt +0 -3
- checkpoint/rank00015.pt +0 -3
- checkpoint/text_projector.pt +0 -3
- config.json +14 -3
- manifest.json +64 -0
- checkpoint/rank00000.pt → model.safetensors +2 -2
- rae.json +212 -0
- rae.pt +0 -3
README.md
CHANGED
|
@@ -1,18 +1,7 @@
|
|
| 1 |
-
|
| 2 |
-
license: cc-by-nc-4.0
|
| 3 |
-
pipeline_tag: image-to-3d
|
| 4 |
-
library_name: oneworld
|
| 5 |
-
tags:
|
| 6 |
-
- 3d-gaussian-splatting
|
| 7 |
-
- image-to-3d
|
| 8 |
-
- multi-view-generation
|
| 9 |
-
---
|
| 10 |
|
| 11 |
-
|
| 12 |
|
| 13 |
-
|
| 14 |
-
It supports image-and-camera inference with optional text conditioning. Use the
|
| 15 |
-
inference code from `SensenGao/OneWorld`.
|
| 16 |
|
| 17 |
-
|
| 18 |
-
statistics. Pi3X and Wan2.1-T2V-1.3B are downloaded separately.
|
|
|
|
| 1 |
+
# OneWorld merged inference weights
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2 |
|
| 3 |
+
One complete `model.safetensors` contains the generator EMA, RAE with final decoder EMA, and text-projector EMA, stored under their original parameter names. All values retain their original precision. There are no rank-local model shards.
|
| 4 |
|
| 5 |
+
Use the matching repaired inference code supporting `oneworld-4step-v3`, with the public Wan2.1-T2V-1.3B and Pi3X backbones. The compatible code is published in [GitHub commit ebd6f31](https://github.com/SensenGao/OneWorld/commit/ebd6f316ce8fe1da74650a9a5d5738e2ec85994a); follow its [setup and inference instructions](https://github.com/SensenGao/OneWorld/blob/ebd6f316ce8fe1da74650a9a5d5738e2ec85994a/infer/README.md). The default HF main release is ready to use directly, without a repair or merge step. No input-alignment initializer or decoder shard-layout file is needed. Latent statistics are stored separately in `latent_stats.pt`.
|
|
|
|
|
|
|
| 6 |
|
| 7 |
+
Verify with `python infer/verify_weights.py --model /path/to/this/package`, then use the same path with `python infer/infer.py --model ...`.
|
|
|
checkpoint/contract.json
CHANGED
|
@@ -1,5 +1,5 @@
|
|
| 1 |
{
|
| 2 |
-
"format": "oneworld-4step-
|
| 3 |
"checkpoint_format": "oneworld-rank-local-inference-v1",
|
| 4 |
"world_size": 16,
|
| 5 |
"step": 30000,
|
|
@@ -30,5 +30,10 @@
|
|
| 30 |
],
|
| 31 |
"rollout_feedback": "3dgs_render_reencode",
|
| 32 |
"cfg": false
|
| 33 |
-
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 34 |
}
|
|
|
|
| 1 |
{
|
| 2 |
+
"format": "oneworld-4step-v3",
|
| 3 |
"checkpoint_format": "oneworld-rank-local-inference-v1",
|
| 4 |
"world_size": 16,
|
| 5 |
"step": 30000,
|
|
|
|
| 30 |
],
|
| 31 |
"rollout_feedback": "3dgs_render_reencode",
|
| 32 |
"cfg": false
|
| 33 |
+
},
|
| 34 |
+
"decoder_layout": "rae_decoder_v1",
|
| 35 |
+
"appearance_posterior": "identity",
|
| 36 |
+
"input_architecture": "aligned",
|
| 37 |
+
"weight_layout": "named_parameters_v1",
|
| 38 |
+
"decoder_ema_merged": true
|
| 39 |
}
|
checkpoint/rank00001.pt
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:7d8575ccbc4efbfb550ff5493e59d83cc01e1d737d79a1075bb91921d9eb6ae3
|
| 3 |
-
size 465166115
|
|
|
|
|
|
|
|
|
|
|
|
checkpoint/rank00002.pt
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:c453cc9f7bc61a91c01481e9db5055a22f88cda283b03938d896258eb4bf1776
|
| 3 |
-
size 465166115
|
|
|
|
|
|
|
|
|
|
|
|
checkpoint/rank00003.pt
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:20c52eff0f7ec256c1aaf8036ed291f977a4e99aad360d4a5dc5818aa74f62cc
|
| 3 |
-
size 465166115
|
|
|
|
|
|
|
|
|
|
|
|
checkpoint/rank00004.pt
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:d28762ae7cf3f8f3da14119780ae054a0f64ce420563a44e0cb879a59754a0c9
|
| 3 |
-
size 465166115
|
|
|
|
|
|
|
|
|
|
|
|
checkpoint/rank00005.pt
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:b9e4cbf150c1d348b96db59c3805dda178335d2113b780a3b2662ab392be6802
|
| 3 |
-
size 465166115
|
|
|
|
|
|
|
|
|
|
|
|
checkpoint/rank00006.pt
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:fa952a75601176253678ef434e11db0ee9562a0ffb62a7be08b4461c16476678
|
| 3 |
-
size 465166115
|
|
|
|
|
|
|
|
|
|
|
|
checkpoint/rank00007.pt
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:6be9c67bacc8baa69c1df4e50df2e6c91da84a4c112c3277673f8b37cd4518ca
|
| 3 |
-
size 465166115
|
|
|
|
|
|
|
|
|
|
|
|
checkpoint/rank00008.pt
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:811fc117730a9aa1659e1de2a30c276e1113da02916f5ca356bb8ca005f3fb47
|
| 3 |
-
size 465166115
|
|
|
|
|
|
|
|
|
|
|
|
checkpoint/rank00009.pt
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:c9235fc4104def92815cd58b456b5e0f0a05bef3c2e45e3922e6eac4c8b08b13
|
| 3 |
-
size 465166115
|
|
|
|
|
|
|
|
|
|
|
|
checkpoint/rank00010.pt
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:77dfef188448c1e72aa9a80242673fcabcb8624df8c03a280de090a1b8fed2cf
|
| 3 |
-
size 465166115
|
|
|
|
|
|
|
|
|
|
|
|
checkpoint/rank00011.pt
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:56ff97861c76d9a3ffa71697fc12ac772e6f2b7312e65fa6d42dfcc206f9a517
|
| 3 |
-
size 465166115
|
|
|
|
|
|
|
|
|
|
|
|
checkpoint/rank00012.pt
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:5015141d5e0d006a07c4a32bf163d61d875f8f5fd9a5c141446ab2411bae11fe
|
| 3 |
-
size 465166115
|
|
|
|
|
|
|
|
|
|
|
|
checkpoint/rank00013.pt
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:130290b0fa1d80010c2060738d49eed66a33ebc84ccc532160cda025be8f379b
|
| 3 |
-
size 465166115
|
|
|
|
|
|
|
|
|
|
|
|
checkpoint/rank00014.pt
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:173b6ab77075e08bb135d2a870e629c0afae69ff8cd1151f9fd1348873a0f4d9
|
| 3 |
-
size 465166115
|
|
|
|
|
|
|
|
|
|
|
|
checkpoint/rank00015.pt
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:f8366886ee075c2745859d4e48205770e974a5aec23397bf01d12914f3644f04
|
| 3 |
-
size 465166115
|
|
|
|
|
|
|
|
|
|
|
|
checkpoint/text_projector.pt
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:98db43b8e9760639ca03ddcae38bcf8ff92e19686a074d55fc3e026c7b05e4b6
|
| 3 |
-
size 69234330
|
|
|
|
|
|
|
|
|
|
|
|
config.json
CHANGED
|
@@ -1,10 +1,21 @@
|
|
| 1 |
{
|
| 2 |
-
"format": "oneworld-4step-
|
| 3 |
"checkpoint": "checkpoint",
|
| 4 |
-
"rae": "rae.
|
| 5 |
"stats": "latent_stats.pt",
|
| 6 |
"text_length": 256,
|
| 7 |
"views": 8,
|
| 8 |
"height": 224,
|
| 9 |
-
"width": 448
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 10 |
}
|
|
|
|
| 1 |
{
|
| 2 |
+
"format": "oneworld-4step-v3",
|
| 3 |
"checkpoint": "checkpoint",
|
| 4 |
+
"rae": "rae.json",
|
| 5 |
"stats": "latent_stats.pt",
|
| 6 |
"text_length": 256,
|
| 7 |
"views": 8,
|
| 8 |
"height": 224,
|
| 9 |
+
"width": 448,
|
| 10 |
+
"aligned_input": true,
|
| 11 |
+
"manifest": "manifest.json",
|
| 12 |
+
"weights": "model.safetensors",
|
| 13 |
+
"weight_layout": "named_parameters_v1",
|
| 14 |
+
"step": 30000,
|
| 15 |
+
"source_world_size": 16,
|
| 16 |
+
"text_projector_fingerprint": [
|
| 17 |
+
8653824.0,
|
| 18 |
+
-374.2581043791723,
|
| 19 |
+
11370.835342871147
|
| 20 |
+
]
|
| 21 |
}
|
manifest.json
ADDED
|
@@ -0,0 +1,64 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"format": "oneworld-release-manifest-v1",
|
| 3 |
+
"files": {
|
| 4 |
+
"README.md": {
|
| 5 |
+
"size": 1060,
|
| 6 |
+
"sha256": "41155439ce54c6fc2ab3cbd60e3631c9b85acc4e1a8f4784a17c992c53f87d43"
|
| 7 |
+
},
|
| 8 |
+
"checkpoint/contract.json": {
|
| 9 |
+
"size": 819,
|
| 10 |
+
"sha256": "0b801f91893196a2f6384cdc4efbff759b0e949e02c81bb907cb17e5ec82a032"
|
| 11 |
+
},
|
| 12 |
+
"config.json": {
|
| 13 |
+
"size": 463,
|
| 14 |
+
"sha256": "f3605739897b199c9c20d249a02dfa28d7aaaa6ab0d6f16564ccf8dbca872fb0"
|
| 15 |
+
},
|
| 16 |
+
"latent_stats.pt": {
|
| 17 |
+
"size": 20088,
|
| 18 |
+
"sha256": "ffef0f1ddd2efdc5037341ca9299d4180b3580b9b950cfa88bb9f49d9c3efcf1"
|
| 19 |
+
},
|
| 20 |
+
"model.safetensors": {
|
| 21 |
+
"size": 7908626480,
|
| 22 |
+
"sha256": "f882c70372994b5ac06fe5b071167affc2911deebc9fe11f576f34ca5b1dbf74"
|
| 23 |
+
},
|
| 24 |
+
"rae.json": {
|
| 25 |
+
"size": 7387,
|
| 26 |
+
"sha256": "681e0c54c2043e91195af35fec692942642fcd4fff84d87545e59d22af32e18d"
|
| 27 |
+
}
|
| 28 |
+
},
|
| 29 |
+
"source_files_sha256": {
|
| 30 |
+
"README.md": "7e183b389051989e1dbc66d2a14f49c6c22bedfa28cd24e12e34aef0970b090f",
|
| 31 |
+
"checkpoint/contract.json": "3b445b2033923e73db6a2924c355aca089bb13c56713703d54d793d5ab711bb4",
|
| 32 |
+
"checkpoint/layout.json": "d3d0d817b22c923b8a5b635d1b2e9deec82849e8678ed4d4208e174b5411c133",
|
| 33 |
+
"checkpoint/rank00000.pt": "aea4c2d203a435a424f578bed7b11dafa405eb2419e2f441e4096545a914106b",
|
| 34 |
+
"checkpoint/rank00001.pt": "7d8575ccbc4efbfb550ff5493e59d83cc01e1d737d79a1075bb91921d9eb6ae3",
|
| 35 |
+
"checkpoint/rank00002.pt": "c453cc9f7bc61a91c01481e9db5055a22f88cda283b03938d896258eb4bf1776",
|
| 36 |
+
"checkpoint/rank00003.pt": "20c52eff0f7ec256c1aaf8036ed291f977a4e99aad360d4a5dc5818aa74f62cc",
|
| 37 |
+
"checkpoint/rank00004.pt": "d28762ae7cf3f8f3da14119780ae054a0f64ce420563a44e0cb879a59754a0c9",
|
| 38 |
+
"checkpoint/rank00005.pt": "b9e4cbf150c1d348b96db59c3805dda178335d2113b780a3b2662ab392be6802",
|
| 39 |
+
"checkpoint/rank00006.pt": "fa952a75601176253678ef434e11db0ee9562a0ffb62a7be08b4461c16476678",
|
| 40 |
+
"checkpoint/rank00007.pt": "6be9c67bacc8baa69c1df4e50df2e6c91da84a4c112c3277673f8b37cd4518ca",
|
| 41 |
+
"checkpoint/rank00008.pt": "811fc117730a9aa1659e1de2a30c276e1113da02916f5ca356bb8ca005f3fb47",
|
| 42 |
+
"checkpoint/rank00009.pt": "c9235fc4104def92815cd58b456b5e0f0a05bef3c2e45e3922e6eac4c8b08b13",
|
| 43 |
+
"checkpoint/rank00010.pt": "77dfef188448c1e72aa9a80242673fcabcb8624df8c03a280de090a1b8fed2cf",
|
| 44 |
+
"checkpoint/rank00011.pt": "56ff97861c76d9a3ffa71697fc12ac772e6f2b7312e65fa6d42dfcc206f9a517",
|
| 45 |
+
"checkpoint/rank00012.pt": "5015141d5e0d006a07c4a32bf163d61d875f8f5fd9a5c141446ab2411bae11fe",
|
| 46 |
+
"checkpoint/rank00013.pt": "130290b0fa1d80010c2060738d49eed66a33ebc84ccc532160cda025be8f379b",
|
| 47 |
+
"checkpoint/rank00014.pt": "173b6ab77075e08bb135d2a870e629c0afae69ff8cd1151f9fd1348873a0f4d9",
|
| 48 |
+
"checkpoint/rank00015.pt": "f8366886ee075c2745859d4e48205770e974a5aec23397bf01d12914f3644f04",
|
| 49 |
+
"checkpoint/text_projector.pt": "801ae7fccc425a38a78aed75d7e1403975d72b0517e2b2995940fce4cf760f02",
|
| 50 |
+
"config.json": "af64edd46543d3cc9ba5aa96274c75ddc329c84c5b0b7f8e279791b020618568",
|
| 51 |
+
"input_alignment.pt": "a8d6dc5b56a15347a2ba9678c3b9716d7d94ce4e83b1e4026276703786c1c2ff",
|
| 52 |
+
"latent_stats.pt": "ffef0f1ddd2efdc5037341ca9299d4180b3580b9b950cfa88bb9f49d9c3efcf1",
|
| 53 |
+
"rae.pt": "28b9fee8cd179d2249c962c7a9dbfa17316fb6c6a6d0f4df1c0f6a4d65f9a4c5"
|
| 54 |
+
},
|
| 55 |
+
"weight_layout": "named_parameters_v1",
|
| 56 |
+
"trained_parameters_changed": false,
|
| 57 |
+
"precision_conversion": false,
|
| 58 |
+
"generator_ema_parameters": 870,
|
| 59 |
+
"decoder_ema_parameters_merged_into_rae": 449,
|
| 60 |
+
"rae_parameters": 754,
|
| 61 |
+
"total_named_parameters": 1628,
|
| 62 |
+
"source_world_size": 16,
|
| 63 |
+
"source_step": 30000
|
| 64 |
+
}
|
checkpoint/rank00000.pt → model.safetensors
RENAMED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:f882c70372994b5ac06fe5b071167affc2911deebc9fe11f576f34ca5b1dbf74
|
| 3 |
+
size 7908626480
|
rae.json
ADDED
|
@@ -0,0 +1,212 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"format": "oneworld-rae-inference-v2",
|
| 3 |
+
"step": 30000,
|
| 4 |
+
"contract": {
|
| 5 |
+
"version": 19,
|
| 6 |
+
"parent_checkpoint": "",
|
| 7 |
+
"resolution_hw": null,
|
| 8 |
+
"resolutions_hw": [
|
| 9 |
+
[
|
| 10 |
+
224,
|
| 11 |
+
448
|
| 12 |
+
],
|
| 13 |
+
[
|
| 14 |
+
448,
|
| 15 |
+
448
|
| 16 |
+
],
|
| 17 |
+
[
|
| 18 |
+
252,
|
| 19 |
+
448
|
| 20 |
+
],
|
| 21 |
+
[
|
| 22 |
+
336,
|
| 23 |
+
448
|
| 24 |
+
]
|
| 25 |
+
],
|
| 26 |
+
"patch_grids_hw": [
|
| 27 |
+
[
|
| 28 |
+
16,
|
| 29 |
+
32
|
| 30 |
+
],
|
| 31 |
+
[
|
| 32 |
+
32,
|
| 33 |
+
32
|
| 34 |
+
],
|
| 35 |
+
[
|
| 36 |
+
18,
|
| 37 |
+
32
|
| 38 |
+
],
|
| 39 |
+
[
|
| 40 |
+
24,
|
| 41 |
+
32
|
| 42 |
+
]
|
| 43 |
+
],
|
| 44 |
+
"context_views": 8,
|
| 45 |
+
"heldout_target_views": 8,
|
| 46 |
+
"boundary_name": "independent_layernorm_stock_pi3_z18_and_rgb32",
|
| 47 |
+
"boundary_block_zero_based": 17,
|
| 48 |
+
"boundary_attention": "global",
|
| 49 |
+
"pre_boundary": "frozen_stock_pi3_blocks_0_to_17_no_rgb_injection",
|
| 50 |
+
"continuation_blocks": [
|
| 51 |
+
18,
|
| 52 |
+
35
|
| 53 |
+
],
|
| 54 |
+
"register_tokens": "not_generated_shared_five_learned_parameters",
|
| 55 |
+
"pose": "all_16_opencv_c2w_ref0_pi3_native_mean_baseline_scale_aug_1",
|
| 56 |
+
"view_condition": "fixed_context_view0_identity_pose_plus_intrinsic_rays;remaining_context_views_permutation_equivariant",
|
| 57 |
+
"decoders": "independent_full_resolution_rgb_dpt_and_gaussian_dpt;pi3_point_head_depth_on_intrinsic_pixel_rays",
|
| 58 |
+
"level_fusion": "independent_four_scale_softmax_gates",
|
| 59 |
+
"level_drop_probability": 0.0,
|
| 60 |
+
"level_drop_start_step": 30000,
|
| 61 |
+
"decoder_noise": {
|
| 62 |
+
"distribution": "post_fusion_layernorm_per_scene_uniform_sigma_iid_gaussian",
|
| 63 |
+
"tau": 0.0,
|
| 64 |
+
"probability": 1.0,
|
| 65 |
+
"start_step": 0,
|
| 66 |
+
"shared_strength_across_views": true
|
| 67 |
+
},
|
| 68 |
+
"nvs": "eight_endpoint_bracketing_context_to_eight_heldout_interior_targets",
|
| 69 |
+
"view_split": "eight_evenly_spaced_context_including_both_endpoints_plus_eight_interior_targets",
|
| 70 |
+
"losses": "balanced_rgb_nvs_mse_plus_lpips_plus_stock_pi3_depth_and_rendered_depth_no_gan",
|
| 71 |
+
"loss_weights": {
|
| 72 |
+
"rgb": 1.0,
|
| 73 |
+
"nvs": 1.0,
|
| 74 |
+
"merged_context_rgb": 1.0,
|
| 75 |
+
"lpips": 0.05,
|
| 76 |
+
"lpips_start": 0,
|
| 77 |
+
"teacher_depth": 0.01,
|
| 78 |
+
"render_depth": 0.0
|
| 79 |
+
},
|
| 80 |
+
"gan": {
|
| 81 |
+
"discriminator": "dino_s8_ks9_bn_spectral_norm",
|
| 82 |
+
"disc_loss": "hinge",
|
| 83 |
+
"generator_loss": "negative_mean_fake_logits",
|
| 84 |
+
"weight": 0.0,
|
| 85 |
+
"adaptive_max": 10000.0,
|
| 86 |
+
"disc_start_step": 3750,
|
| 87 |
+
"generator_start_step": 5000
|
| 88 |
+
},
|
| 89 |
+
"latent_channels": {
|
| 90 |
+
"stock_pi3_z18": 1024,
|
| 91 |
+
"svg_rgb": 32
|
| 92 |
+
},
|
| 93 |
+
"latent_concat_dim": 1056,
|
| 94 |
+
"dit_latent": "concat(layernorm_stock_pi3_z18,layernorm_rgb32);reversible_per_resolution_channel_standardization_deferred_to_dit",
|
| 95 |
+
"decoder_adapter": "layernorm_z18_plus_zero_init_mlp_layernorm_rgb32_to_1024_then_layernorm_then_fixed_channel_denorm",
|
| 96 |
+
"decoder_taps": [
|
| 97 |
+
17,
|
| 98 |
+
35
|
| 99 |
+
],
|
| 100 |
+
"rgb_alignment": "independent_per_token_layernorm_no_affine",
|
| 101 |
+
"post_concat_norm": null,
|
| 102 |
+
"rgb_direct_skip": "scene_shared_drop_p0_layernorm_rgb32_exact_patch14_full_resolution_detail_merger_before_rgb_output_conv_not_rgb_dpt_concat",
|
| 103 |
+
"rgb_trunk_condition": "zero_init_32_to_256_to_1024_fixed_gate_0.1",
|
| 104 |
+
"rgb_branch_mask": "before_alignment_shared_per_multiview_scene",
|
| 105 |
+
"rgb_branch_drop": 0.0,
|
| 106 |
+
"rgb_branch_drop_scope": "scene_shared_synchronized_full_resolution_rgb_and_gaussian_input_mergers;block19_condition_keeps_clean_rgb",
|
| 107 |
+
"rgb_transformer_dropout": 0.0,
|
| 108 |
+
"training_scope": "full",
|
| 109 |
+
"overfit_scene_pool_size": 0,
|
| 110 |
+
"dataset": "mixed_re10k_nvs_refined",
|
| 111 |
+
"data_mix": "exact_global_sample_parity_50pct_nvs_refined_static_50pct_legacy_re10k",
|
| 112 |
+
"nvs_refined_subsets": [
|
| 113 |
+
"ACID",
|
| 114 |
+
"DL3DV",
|
| 115 |
+
"Re10K"
|
| 116 |
+
],
|
| 117 |
+
"source_view_sampling": {
|
| 118 |
+
"legacy_re10k": "ordered_16_views_existing_loader",
|
| 119 |
+
"nvs_refined_acid": "uniform_16_view_window_span_48_160_endpoint_bracketed_8_to_8",
|
| 120 |
+
"nvs_refined_re10k": "uniform_16_view_window_span_48_160_endpoint_bracketed_8_to_8",
|
| 121 |
+
"nvs_refined_dl3dv": "depthsplat_adapted_span_warmup_15_30_to_20_50_over_10k;8_context=temporal_endpoints_plus_camera_center_fps;8_distinct_targets_inside_context_window"
|
| 122 |
+
},
|
| 123 |
+
"resolution_schedule": "global_step_synchronized_shuffled_uniform_cycle",
|
| 124 |
+
"normalization": {
|
| 125 |
+
"z18": "fixed_layernorm_per_token_no_affine",
|
| 126 |
+
"rgb32": "fixed_layernorm_per_token_no_affine",
|
| 127 |
+
"fused_block18_input": "fixed_layernorm_per_token_no_affine",
|
| 128 |
+
"z36_decoder_tap": "trainable_layernorm_per_token"
|
| 129 |
+
},
|
| 130 |
+
"gaussians": {
|
| 131 |
+
"centres": "pi3_depth_on_intrinsic_pixel_rays_no_xy_offset",
|
| 132 |
+
"ray_mode": "z_depth",
|
| 133 |
+
"sh_degree": 1,
|
| 134 |
+
"attribute_channels": 20,
|
| 135 |
+
"rgb_skip": true,
|
| 136 |
+
"merged_context_rgb_weight": 1.0,
|
| 137 |
+
"single_source_diagonal_rgb_weight": 0.0,
|
| 138 |
+
"render_cloud": "all_context_views_merged_for_every_camera",
|
| 139 |
+
"random_background_during_training": false,
|
| 140 |
+
"source_views_kept_per_training_render": 8,
|
| 141 |
+
"source_cloud_dropout_probability": 0.0,
|
| 142 |
+
"training_geometry_oracle": false
|
| 143 |
+
},
|
| 144 |
+
"geometry_teacher": {
|
| 145 |
+
"source": "frozen_stock_pi3_blocks18_to35_and_point_head",
|
| 146 |
+
"direct_depth_weight": 0.01,
|
| 147 |
+
"rendered_context_depth_weight": 0.0,
|
| 148 |
+
"anchor_end_step_exclusive": 0,
|
| 149 |
+
"direct_depth_mode": "scale_invariant_pixel_log_l1"
|
| 150 |
+
},
|
| 151 |
+
"cross_view_depth_consistency": {
|
| 152 |
+
"weight": 0.0,
|
| 153 |
+
"source_pairing": "directed_nearest_camera",
|
| 154 |
+
"pixel_stride": 4,
|
| 155 |
+
"robust_retained_quantile": 0.7
|
| 156 |
+
},
|
| 157 |
+
"wan_alignment": {
|
| 158 |
+
"status": "active",
|
| 159 |
+
"teacher": "frozen_Wan2.1_VAE_16d_deterministic_mean",
|
| 160 |
+
"frame_contract": "each_view_independent_T1_first_frame",
|
| 161 |
+
"teacher_grid": "H_over_8_W_over_8",
|
| 162 |
+
"teacher_to_latent_grid": "fixed_parameter_free_adaptive_area_aggregation_to_H_over_14",
|
| 163 |
+
"latent_projection": "final_layernorm_RGB32_1x1_conv_32_to_16",
|
| 164 |
+
"loss": "VA-VAE_direct_cosine_plus_token_relation_matrix",
|
| 165 |
+
"weight": 0.1,
|
| 166 |
+
"adaptive_gradient_balance": true,
|
| 167 |
+
"cosine_margin": 0.0,
|
| 168 |
+
"distance_margin": 0.0,
|
| 169 |
+
"relation_token_samples": 256,
|
| 170 |
+
"teacher_encode_batch_size": 1
|
| 171 |
+
},
|
| 172 |
+
"rgb_decoder": {
|
| 173 |
+
"dpt_input": "geometry_z18_and_z36_only_no_rgb32_concat",
|
| 174 |
+
"dpt_prefinal_scale": "8x_patch_grid_then_bilinear_to_image",
|
| 175 |
+
"detail_path": "rgb32_hidden0_pixelshuffle_exact_patch14_to_full_resolution_feature32_plus_3x3_refine",
|
| 176 |
+
"fusion": "feature"
|
| 177 |
+
},
|
| 178 |
+
"decoder_denorm": {
|
| 179 |
+
"format": "pi3-v19-ln-to-raw-channel-affine-v1",
|
| 180 |
+
"statistics": "embedded_in_model.safetensors",
|
| 181 |
+
"position": "after_fusion_layernorm_before_pi3_block18",
|
| 182 |
+
"fixed_during_rae_training": true,
|
| 183 |
+
"scale_mean": 0.09068994224071503,
|
| 184 |
+
"scale_std": 0.0018938040593639016,
|
| 185 |
+
"bias_mean": 0.025266744196414948,
|
| 186 |
+
"bias_std": 0.001551253255456686
|
| 187 |
+
},
|
| 188 |
+
"appearance_posterior": "identity"
|
| 189 |
+
},
|
| 190 |
+
"args": {
|
| 191 |
+
"appearance_dim": 32,
|
| 192 |
+
"appearance_hidden_dim": 384,
|
| 193 |
+
"appearance_depth": 6,
|
| 194 |
+
"appearance_heads": 8,
|
| 195 |
+
"appearance_transformer_dropout": 0.0,
|
| 196 |
+
"appearance_drop": 0.0,
|
| 197 |
+
"features": 256,
|
| 198 |
+
"level_drop": 0.0,
|
| 199 |
+
"sh_degree": 1,
|
| 200 |
+
"gs_ray_mode": "z_depth",
|
| 201 |
+
"rgb_detail_hidden_dim": 0,
|
| 202 |
+
"rgb_detail_merge": "feature",
|
| 203 |
+
"rgb_coarse_scale": 1.0,
|
| 204 |
+
"rgb_detail_scale": 1.0
|
| 205 |
+
},
|
| 206 |
+
"tensor_source": {
|
| 207 |
+
"format": "safetensors",
|
| 208 |
+
"file": "model.safetensors",
|
| 209 |
+
"prefix": "rae."
|
| 210 |
+
},
|
| 211 |
+
"decoder_ema_merged": true
|
| 212 |
+
}
|
rae.pt
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:a72c680e620bea3529365a658535562122638f1b26f11542335af4a56f19e2bc
|
| 3 |
-
size 1526571746
|
|
|
|
|
|
|
|
|
|
|
|