Document available checkpoints, provenance checks, and paper-reported metrics

README.md

@@ -14,14 +14,14 @@ datasets:
   - ILSVRC/imagenet-1k
 ---
 
-# FuseReg: Regularizing Layer Fusion Mitigates the Reconstruction-Generation Gap in Representation Autoencoders
+# Regularizing Layer Fusion Closes the Reconstruction-Generation Gap in Representation Autoencoders
 
 Official checkpoints for **FuseReg**. FuseReg replaces heuristic encoder-layer
 fusion in representation autoencoders (RAEs) with training over random subsets of
 encoder layers, so a single pixel decoder reconstructs reliably under any layer-subset
 fusion and pairs better with the diffusion generator.
 
-**Paper:** coming soon · **Code:** coming soon
+**Paper:** coming soon · **Code:** [Hongyang-Du/FuseReg](https://github.com/Hongyang-Du/FuseReg)
 
 ## Repository layout
 
@@ -30,25 +30,47 @@ dinov3-vitl/                      # DINOv3 ViT-L encoder, ImageNet 256x256
   decoder_k23/
     p0.00.safetensors             # reproduced RAEv2 decoder (no layer drop)
     p0.05.safetensors ... p0.95.safetensors
+  decoder_k7/
+    p0.60.safetensors
+    p0.90.safetensors
   ditxl_k23/
     raev2_pdit0.0_ep040.safetensors
     raev2_pdit0.0_ep080.safetensors
     fusereg_pdit0.3_ep040.safetensors ... fusereg_pdit0.9_ep040.safetensors
-eupe-vitb/                        # coming soon
-siglip/                           # coming soon
+eupe-vitb/                        # no checkpoints published yet
+siglip/                           # no checkpoints published yet
 ```
 
-All files contain **EMA weights only** in `safetensors` format (no optimizer,
-discriminator, or encoder weights). Each file's `safetensors` metadata records
-`encoder`, `p`, `epoch`, `step`, and the encoder `layers` the model was trained on.
+The repository currently contains **16 inference checkpoints**: eight DINOv3-L
+k=23 decoders, two DINOv3-L k=7 decoders, and six DiT-XL generators. The k=7
+`p=0.3`, four EUPE, and four SigLIP variants have not been published. A 25-file
+candidate collection is tracked in the checkpoint manifest; only entries in its
+`files` array are available for download.
+
+[Checkpoint manifest](checkpoint-manifest.json) records file size, SHA-256,
+selected EMA state, and the scope of validation. Fourteen pre-existing files have
+been checked against source checkpoint epoch/step and EMA tensor names/shapes.
+The k=23 `p=0.95` source archive also matches the paper provenance SHA-256, and all
+456 EMA tensors equal the published file. Both newly published k=7 decoders were
+checked against Drive CRC32C, round-trip tensor equality, and Hub SHA-256.
+
+All checkpoint files contain **EMA weights only** in `safetensors` format (no optimizer,
+discriminator, or encoder weights). The `safetensors` metadata records available training provenance such as
+`encoder`, drop rate, `epoch`, and `step`; some older files do not include `layers`.
+Use the matching evaluation configuration and checkpoint manifest for fusion layers.
+
+All numerical results below are **paper-reported**, not independently reproduced
+during checkpoint packaging. File integrity and tensor equality checks do not
+validate PSNR, SSIM, rFID, or gFID. Full metric reproduction requires the exact
+ImageNet evaluation split, encoder checkpoints, preprocessing, and latent statistics.
 
 ## DINOv3 ViT-L
 
 ### Pixel decoders (`dinov3-vitl/decoder_k23/`)
 
 ViT decoder (hidden size 1152, 415.6M params), trained for 16 epochs on ImageNet-256
-with pixel, LPIPS, adversarial, and classifier-surrogate losses, over DINOv3-L layers
-1-23 with random layer-drop rate `p`. Reconstruction on ImageNet val (50k) under three
+with pixel, LPIPS, and adversarial losses, over DINOv3-L layers
+1-23 with random layer-drop rate `p`. Paper-reported reconstruction on ImageNet val (50k) under three
 inference-time fusions:
 
 | File | p | feed k=7 PSNR / SSIM / rFID | feed k=23 PSNR / SSIM / rFID | feed l11 PSNR / SSIM / rFID |
@@ -62,6 +84,14 @@ inference-time fusions:
 | `p0.90` | 0.9 | 23.62 / 0.671 / 0.610 | 27.60 / 0.827 / 0.415 | 24.82 / 0.722 / 0.458 |
 | **`p0.95`** (recommended) | 0.95 | **23.77 / 0.678 / 0.604** | 27.52 / 0.826 / 0.421 | **25.13 / 0.735 / 0.455** |
 
+### Pixel decoders trained on k=7 (`dinov3-vitl/decoder_k7/`)
+
+Published variants: `p0.60.safetensors` and `p0.90.safetensors`. Their source
+checkpoint metadata identifies the seven encoder layers as
+`[11, 13, 15, 17, 19, 21, 23]`; both were saved at epoch 16. These are distinct
+from feeding a k=23-trained decoder with the same seven-layer subset. No new
+reconstruction metrics are claimed for the packaged files.
+
 ### DiT-XL generators (`dinov3-vitl/ditxl_k23/`)
 
 Class-conditional ImageNet-256 DiT-XL (hidden size 1440, 875.3M params) generating in
@@ -79,7 +109,7 @@ the k=23 DINOv3-L latent.
 FuseReg DiT-XL generators are trained with random layer-drop rate `p_dit` over the same
 k=23 latent; the full `p_dit` x `p_dec` grid is in the paper appendix (DiT-XL rate sweep).
 
-Pairing a fixed RAEv2 k=23 generator with the FuseReg `p0.95` decoder reduces
+The paper reports that pairing a fixed RAEv2 k=23 generator with the FuseReg `p0.95` decoder reduces
 unguided gFID from 3.01 to 2.21 while keeping guided gFID at 1.25 (internal guidance
 1.78, 50k samples, 50 Euler steps).
 
@@ -94,14 +124,22 @@ state_dict = load_file(path)
 decoder.load_state_dict(state_dict)   # decoder built from the FuseReg codebase
 ```
 
-The encoder is not included; obtain DINOv3 ViT-L from
+The full RAE checkpoint used for k=23 `p=0.00` stores decoder weights under
+`ema` with a `decoder.` prefix. The published file contains that decoder subtree.
+Its source archive does not record output normalization settings; confirm the
+original RAE evaluation path before using it for a numerical baseline comparison.
+The trained drop decoders and the original RAE baseline must each use the
+appropriate output normalization. The optional CLS-surrogate operation in the
+code is an additive last-layer token mean, not a classifier-surrogate loss.
+
+The encoder and latent normalization statistics are not included. Obtain DINOv3 ViT-L from
 [Meta](https://github.com/facebookresearch/dinov3) under the DINOv3 License.
 
 ## Citation
 
 ```bibtex
 @article{du2026fusereg,
-  title   = {FuseReg: Regularizing Layer Fusion Mitigates the Reconstruction-Generation Gap in Representation Autoencoders},
+  title   = {Regularizing Layer Fusion Closes the Reconstruction-Generation Gap in Representation Autoencoders},
   author  = {Du, Hongyang and Ye, Junjie and Yang, Jiawei and Xie, Yunfei and Cong, Xiaoyan and Zhang, Haodong and Huang, Yongchao and Li, Zongxia and Kadav, Asim and Wei, Chen and Balestriero, Randall and Wang, Yue},
   year    = {2026}
 }

checkpoint-manifest.json

@@ -0,0 +1,609 @@
+{
+  "repo_id": "Hongyang-Du/FuseReg",
+  "revision": "3f6f7c58434256c835bad6e79df1b598db5d59a1",
+  "format": "safetensors",
+  "weight_selection": "EMA inference weights only",
+  "note": "Checkpoint integrity checks are not an independent reproduction of paper metrics. Encoder weights and latent normalization statistics are separate dependencies.",
+  "present_count": 16,
+  "planned_count": 25,
+  "blocked_files": [
+    {
+      "path": "dinov3-vitl/decoder_k7/p0.30.safetensors",
+      "status": "blocked",
+      "reason": "Google Drive download quota exceeded; connected Drive raw-file tool has a 256 MiB file limit."
+    },
+    {
+      "path": "eupe-vitb/decoder_k11/p0.90.safetensors",
+      "status": "blocked",
+      "reason": "Google Drive download quota exceeded; connected Drive raw-file tool has a 256 MiB file limit."
+    },
+    {
+      "path": "eupe-vitb/decoder_k11/p0.60.safetensors",
+      "status": "blocked",
+      "reason": "Google Drive download quota exceeded; connected Drive raw-file tool has a 256 MiB file limit."
+    },
+    {
+      "path": "eupe-vitb/decoder_k11/p0.30.safetensors",
+      "status": "blocked",
+      "reason": "Google Drive download quota exceeded; connected Drive raw-file tool has a 256 MiB file limit."
+    },
+    {
+      "path": "eupe-vitb/decoder_k11/p0.00.safetensors",
+      "status": "blocked",
+      "reason": "Google Drive download quota exceeded; connected Drive raw-file tool has a 256 MiB file limit."
+    },
+    {
+      "path": "siglip/decoder_k23/p0.90.safetensors",
+      "status": "blocked",
+      "reason": "Google Drive download quota exceeded; connected Drive raw-file tool has a 256 MiB file limit."
+    },
+    {
+      "path": "siglip/decoder_k23/p0.60.safetensors",
+      "status": "blocked",
+      "reason": "Google Drive download quota exceeded; connected Drive raw-file tool has a 256 MiB file limit."
+    },
+    {
+      "path": "siglip/decoder_k23/p0.30.safetensors",
+      "status": "blocked",
+      "reason": "Google Drive download quota exceeded; connected Drive raw-file tool has a 256 MiB file limit."
+    },
+    {
+      "path": "siglip/decoder_k23/p0.00.safetensors",
+      "status": "blocked",
+      "reason": "Google Drive download quota exceeded; connected Drive raw-file tool has a 256 MiB file limit."
+    }
+  ],
+  "files": [
+    {
+      "path": "dinov3-vitl/decoder_k23/p0.00.safetensors",
+      "size_bytes": 1662643616,
+      "sha256": "6b8cd680a15cfa82e62936bf723dca3d57cbfdb4d7cb5d2b5b7686b901086cf4",
+      "encoder": "dinov3-vitl",
+      "model": "decoder",
+      "p": 0.0,
+      "epoch": 16,
+      "layers": [
+        1,
+        2,
+        3,
+        4,
+        5,
+        6,
+        7,
+        8,
+        9,
+        10,
+        11,
+        12,
+        13,
+        14,
+        15,
+        16,
+        17,
+        18,
+        19,
+        20,
+        21,
+        22,
+        23
+      ],
+      "verification": "Drive archive and Hub headers agree on epoch, step, and all selected EMA tensor names and shapes; weight values not independently compared.",
+      "source_key": "ema.decoder.* (prefix removed)",
+      "layers_verification": "Inferred from source filename and existing repository documentation; absent in safetensors metadata.",
+      "output_normalization": "unresolved: source checkpoint lacks output normalization configuration; confirm original RAE evaluation path before reporting metrics"
+    },
+    {
+      "path": "dinov3-vitl/decoder_k23/p0.05.safetensors",
+      "size_bytes": 1662643712,
+      "sha256": "0d43b5477cefa739160103e371594e279733fadc8fc7c41daa5cbda73161ded2",
+      "encoder": "dinov3-vitl",
+      "model": "decoder",
+      "p": 0.05,
+      "epoch": 16,
+      "layers": [
+        1,
+        2,
+        3,
+        4,
+        5,
+        6,
+        7,
+        8,
+        9,
+        10,
+        11,
+        12,
+        13,
+        14,
+        15,
+        16,
+        17,
+        18,
+        19,
+        20,
+        21,
+        22,
+        23
+      ],
+      "verification": "Drive archive and Hub headers agree on epoch, step, and all selected EMA tensor names and shapes; weight values not independently compared.",
+      "source_key": "ema_dec"
+    },
+    {
+      "path": "dinov3-vitl/decoder_k23/p0.10.safetensors",
+      "size_bytes": 1662643712,
+      "sha256": "8b5f611f6f69b1527ef061214044a93ed5f1a9e7a5e30d07a4b7a38c26c6b00b",
+      "encoder": "dinov3-vitl",
+      "model": "decoder",
+      "p": 0.1,
+      "epoch": 16,
+      "layers": [
+        1,
+        2,
+        3,
+        4,
+        5,
+        6,
+        7,
+        8,
+        9,
+        10,
+        11,
+        12,
+        13,
+        14,
+        15,
+        16,
+        17,
+        18,
+        19,
+        20,
+        21,
+        22,
+        23
+      ],
+      "verification": "Drive archive and Hub headers agree on epoch, step, and all selected EMA tensor names and shapes; weight values not independently compared.",
+      "source_key": "ema_dec"
+    },
+    {
+      "path": "dinov3-vitl/decoder_k23/p0.30.safetensors",
+      "size_bytes": 1662643712,
+      "sha256": "47ea1d8ab595098c7ef2a1639b65907ace9506e0e3e30165de8107f378685fbc",
+      "encoder": "dinov3-vitl",
+      "model": "decoder",
+      "p": 0.3,
+      "epoch": 16,
+      "layers": [
+        1,
+        2,
+        3,
+        4,
+        5,
+        6,
+        7,
+        8,
+        9,
+        10,
+        11,
+        12,
+        13,
+        14,
+        15,
+        16,
+        17,
+        18,
+        19,
+        20,
+        21,
+        22,
+        23
+      ],
+      "verification": "Drive archive and Hub headers agree on epoch, step, and all selected EMA tensor names and shapes; weight values not independently compared.",
+      "source_key": "ema_dec"
+    },
+    {
+      "path": "dinov3-vitl/decoder_k23/p0.50.safetensors",
+      "size_bytes": 1662643712,
+      "sha256": "4fba21d3a8ae23df7d5ea7bd143de00f885d562678b50521a11b132321f3191a",
+      "encoder": "dinov3-vitl",
+      "model": "decoder",
+      "p": 0.5,
+      "epoch": 16,
+      "layers": [
+        1,
+        2,
+        3,
+        4,
+        5,
+        6,
+        7,
+        8,
+        9,
+        10,
+        11,
+        12,
+        13,
+        14,
+        15,
+        16,
+        17,
+        18,
+        19,
+        20,
+        21,
+        22,
+        23
+      ],
+      "verification": "Drive archive and Hub headers agree on epoch, step, and all selected EMA tensor names and shapes; weight values not independently compared.",
+      "source_key": "ema_dec"
+    },
+    {
+      "path": "dinov3-vitl/decoder_k23/p0.70.safetensors",
+      "size_bytes": 1662643712,
+      "sha256": "330f76d9775065adee902a5657656af23be13496f6ef97cdf5abba240ba0a253",
+      "encoder": "dinov3-vitl",
+      "model": "decoder",
+      "p": 0.7,
+      "epoch": 16,
+      "layers": [
+        1,
+        2,
+        3,
+        4,
+        5,
+        6,
+        7,
+        8,
+        9,
+        10,
+        11,
+        12,
+        13,
+        14,
+        15,
+        16,
+        17,
+        18,
+        19,
+        20,
+        21,
+        22,
+        23
+      ],
+      "verification": "Drive archive and Hub headers agree on epoch, step, and all selected EMA tensor names and shapes; weight values not independently compared.",
+      "source_key": "ema_dec"
+    },
+    {
+      "path": "dinov3-vitl/decoder_k23/p0.90.safetensors",
+      "size_bytes": 1662643712,
+      "sha256": "042190ff72f3606dc8c1431878f71fc7aeb73275e698bc32b8708066521fbfa7",
+      "encoder": "dinov3-vitl",
+      "model": "decoder",
+      "p": 0.9,
+      "epoch": 16,
+      "layers": [
+        1,
+        2,
+        3,
+        4,
+        5,
+        6,
+        7,
+        8,
+        9,
+        10,
+        11,
+        12,
+        13,
+        14,
+        15,
+        16,
+        17,
+        18,
+        19,
+        20,
+        21,
+        22,
+        23
+      ],
+      "verification": "Drive archive and Hub headers agree on epoch, step, and all selected EMA tensor names and shapes; weight values not independently compared.",
+      "source_key": "ema_dec"
+    },
+    {
+      "path": "dinov3-vitl/decoder_k23/p0.95.safetensors",
+      "size_bytes": 1662643664,
+      "sha256": "f82adb05ca36c5bc22630a86c7439a630769bc2dd7c8f2a5e4043f0e38120476",
+      "encoder": "dinov3-vitl",
+      "model": "decoder",
+      "p": 0.95,
+      "epoch": 16,
+      "layers": [
+        1,
+        2,
+        3,
+        4,
+        5,
+        6,
+        7,
+        8,
+        9,
+        10,
+        11,
+        12,
+        13,
+        14,
+        15,
+        16,
+        17,
+        18,
+        19,
+        20,
+        21,
+        22,
+        23
+      ],
+      "verification": "Source archive SHA-256 matches paper provenance; all 456 EMA tensors equal existing safetensors; local safetensors SHA-256 equals Hub LFS SHA-256.",
+      "source_key": "ema_dec",
+      "source_sha256": "49cf7f50310dd836f9d8282966c4ea5c188fbf03cbfe5504a09efe5fcec79a24"
+    },
+    {
+      "path": "dinov3-vitl/decoder_k7/p0.60.safetensors",
+      "size_bytes": 1662643816,
+      "sha256": "1b627589651d9755ad974f62fe0c54b159016e1a6803a931c1fb0b93d17566d1",
+      "encoder": "dinov3-vitl",
+      "model": "decoder",
+      "p": 0.6,
+      "epoch": 16,
+      "layers": [
+        11,
+        13,
+        15,
+        17,
+        19,
+        21,
+        23
+      ],
+      "source_sha256": "adfcae18e0059d98274d60d017b286f09e3681b582cefc190a15edebd58d875b",
+      "verification": "Drive source size and CRC32C verified; EMA tensors round-trip equal; uploaded size and SHA-256 match Hub LFS metadata."
+    },
+    {
+      "path": "dinov3-vitl/decoder_k7/p0.90.safetensors",
+      "size_bytes": 1662643816,
+      "sha256": "25e0ceed420efdb957246e406e2a623974b55f2b8e1bccfdc9ac69d6eeeb6071",
+      "encoder": "dinov3-vitl",
+      "model": "decoder",
+      "p": 0.9,
+      "epoch": 16,
+      "layers": [
+        11,
+        13,
+        15,
+        17,
+        19,
+        21,
+        23
+      ],
+      "source_sha256": "6110c7e689703215c19feefeac1dd40db64949fb3c863b9ca1c7f7ccb510160f",
+      "verification": "Drive source size and CRC32C verified; EMA tensors round-trip equal; uploaded size and SHA-256 match Hub LFS metadata."
+    },
+    {
+      "path": "dinov3-vitl/ditxl_k23/fusereg_pdit0.3_ep040.safetensors",
+      "size_bytes": 3501338104,
+      "sha256": "303c28286c6d97aea769cbf88fc767828dd52b6da9ea77f50f40b392cb70cd35",
+      "encoder": "dinov3-vitl",
+      "model": "dit-xl",
+      "p": 0.3,
+      "epoch": 40,
+      "layers": [
+        1,
+        2,
+        3,
+        4,
+        5,
+        6,
+        7,
+        8,
+        9,
+        10,
+        11,
+        12,
+        13,
+        14,
+        15,
+        16,
+        17,
+        18,
+        19,
+        20,
+        21,
+        22,
+        23
+      ],
+      "verification": "Drive archive and Hub headers agree on epoch, step, and all selected EMA tensor names and shapes; weight values not independently compared.",
+      "source_key": "ema",
+      "layers_verification": "Inferred from source filename and existing repository documentation; absent in safetensors metadata."
+    },
+    {
+      "path": "dinov3-vitl/ditxl_k23/fusereg_pdit0.5_ep040.safetensors",
+      "size_bytes": 3501338104,
+      "sha256": "e92e22a1f3fc1b00687d0172fabe123b197c714cbfdb39647cf3cdfaa5a4f94f",
+      "encoder": "dinov3-vitl",
+      "model": "dit-xl",
+      "p": 0.5,
+      "epoch": 40,
+      "layers": [
+        1,
+        2,
+        3,
+        4,
+        5,
+        6,
+        7,
+        8,
+        9,
+        10,
+        11,
+        12,
+        13,
+        14,
+        15,
+        16,
+        17,
+        18,
+        19,
+        20,
+        21,
+        22,
+        23
+      ],
+      "verification": "Drive archive and Hub headers agree on epoch, step, and all selected EMA tensor names and shapes; weight values not independently compared.",
+      "source_key": "ema",
+      "layers_verification": "Inferred from source filename and existing repository documentation; absent in safetensors metadata."
+    },
+    {
+      "path": "dinov3-vitl/ditxl_k23/fusereg_pdit0.7_ep040.safetensors",
+      "size_bytes": 3501338104,
+      "sha256": "48124b905b5040f3e78e6eeb5bfe7aa233bf07518392b0cfe9344fa65a937797",
+      "encoder": "dinov3-vitl",
+      "model": "dit-xl",
+      "p": 0.7,
+      "epoch": 40,
+      "layers": [
+        1,
+        2,
+        3,
+        4,
+        5,
+        6,
+        7,
+        8,
+        9,
+        10,
+        11,
+        12,
+        13,
+        14,
+        15,
+        16,
+        17,
+        18,
+        19,
+        20,
+        21,
+        22,
+        23
+      ],
+      "verification": "Drive archive and Hub headers agree on epoch, step, and all selected EMA tensor names and shapes; weight values not independently compared.",
+      "source_key": "ema",
+      "layers_verification": "Inferred from source filename and existing repository documentation; absent in safetensors metadata."
+    },
+    {
+      "path": "dinov3-vitl/ditxl_k23/fusereg_pdit0.9_ep040.safetensors",
+      "size_bytes": 3501338104,
+      "sha256": "10d1b10d343273de043f43ed1f7bb93ebc58c462ef20c5bbf610f86d6e252ce8",
+      "encoder": "dinov3-vitl",
+      "model": "dit-xl",
+      "p": 0.9,
+      "epoch": 40,
+      "layers": [
+        1,
+        2,
+        3,
+        4,
+        5,
+        6,
+        7,
+        8,
+        9,
+        10,
+        11,
+        12,
+        13,
+        14,
+        15,
+        16,
+        17,
+        18,
+        19,
+        20,
+        21,
+        22,
+        23
+      ],
+      "verification": "Drive archive and Hub headers agree on epoch, step, and all selected EMA tensor names and shapes; weight values not independently compared.",
+      "source_key": "ema",
+      "layers_verification": "Inferred from source filename and existing repository documentation; absent in safetensors metadata."
+    },
+    {
+      "path": "dinov3-vitl/ditxl_k23/raev2_pdit0.0_ep040.safetensors",
+      "size_bytes": 3501338168,
+      "sha256": "47090a0e892af0aa60fb9961f9fbf38873c2dd5b940eb08a5028ad8450e4fd62",
+      "encoder": "dinov3-vitl",
+      "model": "dit-xl",
+      "p": 0.0,
+      "epoch": 40,
+      "layers": [
+        1,
+        2,
+        3,
+        4,
+        5,
+        6,
+        7,
+        8,
+        9,
+        10,
+        11,
+        12,
+        13,
+        14,
+        15,
+        16,
+        17,
+        18,
+        19,
+        20,
+        21,
+        22,
+        23
+      ],
+      "verification": "Drive archive and Hub headers agree on epoch, step, and all selected EMA tensor names and shapes; weight values not independently compared.",
+      "source_key": "ema"
+    },
+    {
+      "path": "dinov3-vitl/ditxl_k23/raev2_pdit0.0_ep080.safetensors",
+      "size_bytes": 3501338168,
+      "sha256": "e07d5ab7e1479a6f0dc9b3ed1ced797970b36a9d9c1a12c36b0c62719b03d074",
+      "encoder": "dinov3-vitl",
+      "model": "dit-xl",
+      "p": 0.0,
+      "epoch": 80,
+      "layers": [
+        1,
+        2,
+        3,
+        4,
+        5,
+        6,
+        7,
+        8,
+        9,
+        10,
+        11,
+        12,
+        13,
+        14,
+        15,
+        16,
+        17,
+        18,
+        19,
+        20,
+        21,
+        22,
+        23
+      ],
+      "verification": "Drive archive and Hub headers agree on epoch, step, and all selected EMA tensor names and shapes; weight values not independently compared.",
+      "source_key": "ema"
+    }
+  ]
+}

dinov3-vitl/README.md

@@ -1,6 +1,7 @@
 # DINOv3 ViT-L
 
-- `decoder_k23/`: FuseReg pixel decoders over DINOv3-L layers 1-23, layer-drop rate `p` in {0, 0.05, 0.1, 0.3, 0.5, 0.7, 0.9, 0.95}. `p0.00` is the reproduced RAEv2 decoder.
-- `ditxl_k23/`: class-conditional ImageNet-256 DiT-XL generators in the k=23 latent (RAEv2 `p_dit=0` at 40/80 epochs; FuseReg `p_dit` in {0.3, 0.5, 0.7, 0.9} at 40 epochs).
+- `decoder_k23/`: eight pixel decoders with `p` in {0, 0.05, 0.1, 0.3, 0.5, 0.7, 0.9, 0.95}.
+- `decoder_k7/`: pixel decoders with `p` in {0.6, 0.9}, trained on layers [11, 13, 15, 17, 19, 21, 23]. The p=0.3 variant is not published.
+- `ditxl_k23/`: six class-conditional DiT-XL generators (RAEv2 p=0 at epochs 40/80; FuseReg p in {0.3, 0.5, 0.7, 0.9} at epoch 40).
 
-See the top-level README for metrics and usage.
+All files contain EMA inference weights only. See the top-level README and checkpoint manifest for validation scope, baseline normalization caveats, and paper-reported results. Latent statistics and encoder weights are separate dependencies.

eupe-vitb/README.md

@@ -1,3 +1,3 @@
 # EUPE ViT-B
 
-Coming soon: FuseReg pixel decoders over EUPE ViT-B layers 1-11 (`decoder_k11/`).
+No EUPE checkpoints have been published in this repository yet.

siglip/README.md

@@ -1,3 +1,3 @@
 # SigLIP
 
-Coming soon: FuseReg pixel decoders over SigLIP layers 1-23 (`decoder_k23/`).
+No SigLIP checkpoints have been published in this repository yet.