Upload LoRA adapter — DIODE Indoors

README.md

@@ -0,0 +1,71 @@
+---
+library_name: peft
+tags:
+  - depth-estimation
+  - lora
+  - peft
+  - metric3d
+  - diode
+  - indoor
+license: apache-2.0
+---
+
+# Metric3D ViT-giant2 — LoRA adapter (DIODE Indoors)
+
+> **Base model:** [`metric3d_vit_giant2`](https://github.com/YvanYin/Metric3D)
+> loaded via `torch.hub.load("yvanyin/metric3d", "metric3d_vit_giant2")`
+
+LoRA fine-tuning of [Metric3D v2](https://github.com/YvanYin/Metric3D)
+(`metric3d_vit_giant2`) on the [DIODE](https://diode-dataset.org/) indoor split (metric depth in metres).
+
+## Training details
+
+| | |
+|---|---|
+| Base model | `metric3d_vit_giant2` (torch.hub `yvanyin/metric3d`) |
+| Dataset | the [DIODE](https://diode-dataset.org/) indoor split (metric depth in metres) |
+| Loss | Direct metric-depth L1 + gradient loss |
+| LoRA rank / alpha | 16 / 32 |
+| LoRA targets | `qkv`, `proj` |
+
+## Usage
+
+```python
+import torch
+from peft import PeftModel
+import torch.nn as nn
+import torch.utils.checkpoint as torch_checkpoint
+
+# Load base model
+model = torch.hub.load("yvanyin/metric3d", "metric3d_vit_giant2",
+                       pretrain=True, trust_repo=True)
+
+# Apply the same gradient-checkpointing wrapper used during training
+# (needed so PEFT key names match the saved adapter)
+def enable_gradient_checkpointing(model):
+    try:
+        encoder = model.depth_model.encoder
+    except AttributeError:
+        encoder = model.base_model.model.depth_model.encoder
+
+    class _CheckpointedBlock(nn.Module):
+        def __init__(self, block):
+            super().__init__()
+            self.block = block
+        def forward(self, x):
+            return torch_checkpoint.checkpoint(self.block, x, use_reentrant=False)
+
+    for blk_group in encoder.blocks:
+        for key in list(blk_group._modules.keys()):
+            blk_group._modules[key] = _CheckpointedBlock(blk_group._modules[key])
+
+enable_gradient_checkpointing(model)
+model = PeftModel.from_pretrained(model, "igzi/depth-lora-checkpoints_diode-diode_indoors")
+model.eval()
+
+# Inference: input pixel_values shape (B, 3, 616, 1064), values normalised
+# with mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375]
+pred_canonical, _, _ = model({"input": pixel_values})
+# De-canonicalise: pred_metric = pred_canonical * (fx_scaled / 1000)
+```
+

adapter_config.json

@@ -0,0 +1,44 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "DepthModel",
+    "parent_library": "mono.model.monodepth_model"
+  },
+  "base_model_name_or_path": null,
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "qkv",
+    "proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}

adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3b126cc053e0cfbe858bd69f34bbf0db0cc7362a1c24324cf610a89811cb26d5
+size 23752944