Upload folder using huggingface_hub

README.md

@@ -1,99 +1,73 @@
 ---
-language:
-- bo
 license: apache-2.0
-library_name: transformers
 tags:
-- image-classification
-- dinov3
-- tibetan
-- manuscript
-- binary-classification
-- vision
-datasets:
-- openpecha/uchen-ume-classification-benchmark
-metrics:
-- accuracy
-- f1
-- auc_roc
-base_model: facebook/dinov3-vits16-pretrain-lvd1689m
+  - image-classification
+  - tibetan
+  - uchen
+  - ume
+library_name: transformers
+pipeline_tag: image-classification
 ---
 
-# Tibetan Script Router (DINOv3-ViT-S)
+# Uchen vs Umê classifier (DINOv3 ViT-S)
 
-This model is a fine-tuned version of **Meta's DINOv3-ViT-S/16** specifically designed for high-precision binary classification of Tibetan scripts. It acts as the primary "Router" in a hierarchical classification pipeline, distinguishing between formal block scripts (**Uchen**) and cursive families (**Ume**).
+Binary Tibetan script classifier: **uchen** (printed) vs **ume** (cursive).
 
-##  Model Details
+## Recommended weights
 
-- **Project Name:** The BDRC Etext Corpus
-- **Developed by:** Dharmaduta
-- **Specifications provided by:** [Buddhist Digital Resource Center (BDRC)](https://www.bdrc.io)
-- **Funded by:** Khyentse Foundation
-- **Model type:** Vision Transformer (ViT)
-- **License:** Apache 2.0
-- **Fine-tuned from:** `facebook/dinov3-vits16-pretrain-lvd1689m`
+Use **`without_preprocess/final_model.pt`** for **full manuscript pages** (no center crop at inference).
 
-##  Dataset & Class Distribution
+| Variant | Preprocess at train | Test F1 | Benchmark F1 |
+|---------|---------------------|---------|----------------|
+| `with_preprocess/` | center_crop train/val, none on test | 0.506 | n/a |
+| `without_preprocess/` | no runtime preprocess | 0.708 | 0.848 |
 
-The model was trained using the [openpecha/uchen-ume-classification](https://huggingface.co/datasets/openpecha/uchen-ume-classification) dataset. This training set consists of **4,572 images** balanced across two major categories. 
+## Benchmark evaluation (held-out 60 pages)
 
-The binary classes were mapped from the following granular script types:
+The benchmark set is **disjoint** from train/val/test. After downloading this repo:
 
-### 1. Uchen (Class 0) — 2,286 Total Samples
-| Granular Script Type | Sample Count |
-| :--- | :--- |
-| `uchen_sugdring` | 1,670 |
-| `uchen_sugthung` | 616 |
+```bash
+pip install torch transformers pillow huggingface_hub scikit-learn
 
-### 2. Ume (Class 1) — 2,286 Total Samples
-| Granular Script Type | Sample Count |
-| :--- | :--- |
-| `petsuk` | 1,388 |
-| `tsegdrig` | 749 |
-| `peri` | 614 |
-| `druthung` | 207 |
-| `tsumachug` | 178 |
-| `yigchung` | 166 |
-| `drudring` | 132 |
-| `drathung` | 129 |
-| `druring` | 119 |
-| `khyuyig` | 113 |
-| `dhumri` | 98 |
-| `tsugchung` | 77 |
-| `trinyig` | 42 |
+# From the dataset repo (has benchmark/ images + inference_uchen_ume.py):
+python inference_uchen_ume.py \
+  --benchmark-dir benchmark \
+  --model-repo openpecha/uchen-ume-classifier \
+  --weights without_preprocess/final_model.pt \
+  --preprocess none
+```
 
-*Note: Classes labeled "Difficult," "Multi-script," and "Non-Tibetan" were excluded to maintain a clean training signal for the Uchen/Ume boundary.*
+Or from this training codebase:
 
-##  Performance Summary
-The model achieved its peak performance at **Stage B** (Partial backbone unfreezing of the last 2 blocks).
+```bash
+python experiments/uchen_ume_binary/eval_benchmark.py \
+  --checkpoint hf_upload/model/without_preprocess/final_model.pt \
+  --benchmark-dir benchmark/benchmark
+```
 
-- **Test Accuracy:** 98.95%
-- **Macro F1-Score:** 0.984
-- **AUC-ROC:** 0.9988
+Reference benchmark run (without_preprocess): **acc 85.0%**, **macro F1 0.848** (30 uchen + 30 ume, full pages, no crop).
 
-### Confusion Matrix
-| Predicted \ Actual | Uchen | Ume |
-|--------------------|-------|-----|
-| **Uchen** | 159   | 2   |
-| **Ume** | 6     | 595 |
-
-##  How to Get Started
+## Load in Python
 
 ```python
-from transformers import AutoImageProcessor, AutoModelForImageClassification
 import torch
+from huggingface_hub import hf_hub_download
+from transformers import AutoImageProcessor
 from PIL import Image
 
-# Note: Gated access approval for DINOv3 is required
-model_id = "openpecha/uchen-ume-classifier"
-processor = AutoImageProcessor.from_pretrained(model_id)
-model = AutoModelForImageClassification.from_pretrained(model_id)
+# See dataset repo inference_uchen_ume.py for DINOv3Classifier + predict
+path = hf_hub_download(
+    "openpecha/uchen-ume-classifier",
+    "without_preprocess/final_model.pt",
+    repo_type="model",
+)
+ckpt = torch.load(path, map_location="cpu", weights_only=False)
+```
+
+## Do not use `with_preprocess/` on full pages
 
-image = Image.open("manuscript_page.jpg").convert("RGB")
-inputs = processor(images=image, return_tensors="pt")
+That variant was trained with **center-crop** on train/val; test F1 on full pages is ~0.51. Only use it with `--preprocess center_crop_whole_page`.
 
-with torch.no_grad():
-    outputs = model(**inputs)
-    prediction = outputs.logits.argmax(-1).item()
+## Training
 
-print(f"Detected Script: {model.config.id2label[prediction]}")
+Backbone: `facebook/dinov3-vits16-pretrain-lvd1689m`. Progressive unfreeze (stages A/B/C). Dataset: [openpecha/uchen-ume-classification-benchmark](https://huggingface.co/datasets/openpecha/uchen-ume-classification-benchmark).

with_preprocess/best_checkpoint.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7646471471367b77ed76fa52ab119075136d00ccdb8d1770c349e7b7e9998196
+size 86674972

with_preprocess/best_checkpoint_name.txt

@@ -0,0 +1 @@
+best_stage_c_last_blocks.pt

with_preprocess/classification_report.txt

@@ -0,0 +1,8 @@
+              precision    recall  f1-score   support
+
+       uchen       0.21      1.00      0.34        99
+         ume       1.00      0.50      0.67       768
+
+    accuracy                           0.56       867
+   macro avg       0.60      0.75      0.51       867
+weighted avg       0.91      0.56      0.63       867

with_preprocess/confusion_matrix.json

@@ -0,0 +1,16 @@
+{
+  "labels": [
+    "uchen",
+    "ume"
+  ],
+  "matrix": [
+    [
+      99,
+      0
+    ],
+    [
+      381,
+      387
+    ]
+  ]
+}

with_preprocess/final_model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e11aee6ea8fd2bbe0090c384580d19fbcd74d66b667ac68e9ad1481c30c9fd70
+size 86672201

with_preprocess/model_card.json

@@ -0,0 +1,15 @@
+{
+  "variant": "with_preprocess",
+  "experiment": "uchen_ume_whole_page",
+  "best_checkpoint": "best_stage_c_last_blocks.pt",
+  "val_macro_f1": 0.9938033069400111,
+  "val_accuracy": 0.9970414201183432,
+  "epoch": 9,
+  "test_metrics": {
+    "loss": 1.5028612467717066,
+    "accuracy": 0.5605536332179931,
+    "macro_f1": 0.5060493910234842,
+    "weighted_f1": 0.6326582036211453,
+    "auc_roc": 0.9685921717171717
+  }
+}

with_preprocess/results.json

@@ -0,0 +1,17 @@
+{
+  "experiment": "uchen_ume_whole_page",
+  "stage_run": "test",
+  "test_metrics": {
+    "loss": 1.5028612467717066,
+    "accuracy": 0.5605536332179931,
+    "macro_f1": 0.5060493910234842,
+    "weighted_f1": 0.6326582036211453,
+    "auc_roc": 0.9685921717171717
+  },
+  "history": {},
+  "report": "              precision    recall  f1-score   support\n\n       uchen       0.21      1.00      0.34        99\n         ume       1.00      0.50      0.67       768\n\n    accuracy                           0.56       867\n   macro avg       0.60      0.75      0.51       867\nweighted avg       0.91      0.56      0.63       867\n",
+  "splits_file": "/root/script-classification-model-train/experiments/uchen_ume_binary/checkpoints/uchen_ume_whole_page/splits.json",
+  "skip_stage_c": false,
+  "stage_c_skip_reason": null,
+  "best_checkpoint": "best_stage_c_last_blocks.pt"
+}

without_preprocess/benchmark_classification_report.txt

@@ -0,0 +1,8 @@
+              precision    recall  f1-score   support
+
+       uchen       0.78      0.97      0.87        30
+         ume       0.96      0.73      0.83        30
+
+    accuracy                           0.85        60
+   macro avg       0.87      0.85      0.85        60
+weighted avg       0.87      0.85      0.85        60

without_preprocess/benchmark_eval_results.json

@@ -0,0 +1,386 @@
+{
+  "checkpoint": "/root/script-classification-model-train/hf_upload/model/without_preprocess/final_model.pt",
+  "benchmark_dir": "/root/script-classification-model-train/benchmark/benchmark",
+  "n_images": 60,
+  "preprocess": "none",
+  "metrics": {
+    "loss": 0.3956117908159892,
+    "accuracy": 0.85,
+    "macro_f1": 0.847930160518164,
+    "weighted_f1": 0.847930160518164,
+    "auc_roc": 0.97
+  },
+  "report": "              precision    recall  f1-score   support\n\n       uchen       0.78      0.97      0.87        30\n         ume       0.96      0.73      0.83        30\n\n    accuracy                           0.85        60\n   macro avg       0.87      0.85      0.85        60\nweighted avg       0.87      0.85      0.85        60\n",
+  "confusion_matrix": [
+    [
+      29,
+      1
+    ],
+    [
+      8,
+      22
+    ]
+  ],
+  "predictions": [
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/uchen/W00KG0555-I1KG1104880320.jpg",
+      "label": "uchen",
+      "pred": "uchen",
+      "prob_ume": 4.636118600132022e-09
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/uchen/W1GS66367-I1GS663690005.jpg",
+      "label": "uchen",
+      "pred": "uchen",
+      "prob_ume": 3.902857861248776e-05
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/uchen/W1PD153537-I1KG131310005.jpg",
+      "label": "uchen",
+      "pred": "uchen",
+      "prob_ume": 0.0006930792005732656
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/uchen/W1PD89084-I1KG134060005.jpg",
+      "label": "uchen",
+      "pred": "uchen",
+      "prob_ume": 6.966766704863403e-06
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/uchen/W23768-26400446.jpg",
+      "label": "uchen",
+      "pred": "uchen",
+      "prob_ume": 0.001230051158927381
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/uchen/W2PD16917-I3PD5910005.jpg",
+      "label": "uchen",
+      "pred": "uchen",
+      "prob_ume": 0.0011563787702471018
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/uchen/W2PD17514-I4PD22210005.jpg",
+      "label": "uchen",
+      "pred": "uchen",
+      "prob_ume": 4.242213981342502e-05
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/uchen/W2PD17517-I4PD15200005.jpg",
+      "label": "uchen",
+      "pred": "uchen",
+      "prob_ume": 8.787653496256098e-05
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/uchen/W2PD19474-I2PD198170005.jpg",
+      "label": "uchen",
+      "pred": "ume",
+      "prob_ume": 0.652681291103363
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/uchen/W2PD20866-I4PD50820480.jpg",
+      "label": "uchen",
+      "pred": "uchen",
+      "prob_ume": 0.01092884037643671
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/uchen/W3CN21390-I3CN222550962.jpg",
+      "label": "uchen",
+      "pred": "uchen",
+      "prob_ume": 0.0006489027291536331
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/uchen/W3CN21414-I2KG2203010005.jpg",
+      "label": "uchen",
+      "pred": "uchen",
+      "prob_ume": 4.1007406252902e-06
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/uchen/W3CN21482-I4CN121200005.jpg",
+      "label": "uchen",
+      "pred": "uchen",
+      "prob_ume": 3.8675672840327024e-05
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/uchen/W3CN27530-I4CN129320548.jpg",
+      "label": "uchen",
+      "pred": "uchen",
+      "prob_ume": 0.007052543107420206
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/uchen/W3CN4180-I3CN41900005.jpg",
+      "label": "uchen",
+      "pred": "uchen",
+      "prob_ume": 0.1726490557193756
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/uchen/W3CN766-I3CN7680053.jpg",
+      "label": "uchen",
+      "pred": "uchen",
+      "prob_ume": 0.0010180854005739093
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/uchen/W3CN8329-I3CN83400005.jpg",
+      "label": "uchen",
+      "pred": "uchen",
+      "prob_ume": 0.0019903830252587795
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/uchen/W3MS261-I3MS3570044.jpg",
+      "label": "uchen",
+      "pred": "uchen",
+      "prob_ume": 0.0010262312134727836
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/uchen/W3MS701-I3MS7080356.jpg",
+      "label": "uchen",
+      "pred": "uchen",
+      "prob_ume": 6.8632389229605906e-06
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/uchen/W3PD885-I3PD9130186.jpg",
+      "label": "uchen",
+      "pred": "uchen",
+      "prob_ume": 1.0850219041458331e-05
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/uchen/W3PD988-I3PD13200005.jpg",
+      "label": "uchen",
+      "pred": "uchen",
+      "prob_ume": 0.005023940000683069
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/uchen/W4CZ58520-I4CZ751270005.jpg",
+      "label": "uchen",
+      "pred": "uchen",
+      "prob_ume": 0.0084664486348629
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/uchen/W4CZ74080-I4CZ741090231.jpg",
+      "label": "uchen",
+      "pred": "uchen",
+      "prob_ume": 8.514942351212085e-08
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/uchen/W4PD1207-I4PD12980005.jpg",
+      "label": "uchen",
+      "pred": "uchen",
+      "prob_ume": 1.6211478826022585e-09
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/uchen/W4PD2050-I4PD20570005.jpg",
+      "label": "uchen",
+      "pred": "uchen",
+      "prob_ume": 0.0007123054238036275
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/uchen/W4PD294-I4PD4110418.jpg",
+      "label": "uchen",
+      "pred": "uchen",
+      "prob_ume": 9.038657822202367e-07
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/uchen/W4PD3075-I4PD31350005.jpg",
+      "label": "uchen",
+      "pred": "uchen",
+      "prob_ume": 7.926726539153606e-05
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/uchen/W4PD3076-I4PD30840005.jpg",
+      "label": "uchen",
+      "pred": "uchen",
+      "prob_ume": 0.00010477022442501038
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/uchen/W8LS19724-I8LS197260350.jpg",
+      "label": "uchen",
+      "pred": "uchen",
+      "prob_ume": 0.001046415651217103
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/uchen/W8LS32739-I8LS339710642.jpg",
+      "label": "uchen",
+      "pred": "uchen",
+      "prob_ume": 2.5808612917899154e-05
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/ume/W1CZ1276-I1CZ17730007.png",
+      "label": "ume",
+      "pred": "ume",
+      "prob_ume": 0.9111959338188171
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/ume/W1CZ2157-I1CZ22190024.jpg",
+      "label": "ume",
+      "pred": "ume",
+      "prob_ume": 0.9920614957809448
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/ume/W1KG22576-I1KG225850005.jpg",
+      "label": "ume",
+      "pred": "uchen",
+      "prob_ume": 0.2963232696056366
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/ume/W1KG4616-I1KG48070506.jpg",
+      "label": "ume",
+      "pred": "ume",
+      "prob_ume": 0.9999186992645264
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/ume/W21872-62960005.jpg",
+      "label": "ume",
+      "pred": "ume",
+      "prob_ume": 0.5267542600631714
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/ume/W23751-I01JW1650259.png",
+      "label": "ume",
+      "pred": "ume",
+      "prob_ume": 0.9908739328384399
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/ume/W24012-36670120.jpg",
+      "label": "ume",
+      "pred": "ume",
+      "prob_ume": 0.9999780654907227
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/ume/W2CZ7987-I1KG38730018.jpg",
+      "label": "ume",
+      "pred": "uchen",
+      "prob_ume": 0.334159791469574
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/ume/W2PD17458-I4PD7310005.jpg",
+      "label": "ume",
+      "pred": "ume",
+      "prob_ume": 0.9986664056777954
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/ume/W2PD17471-I4PD7150594.jpg",
+      "label": "ume",
+      "pred": "ume",
+      "prob_ume": 0.9989309906959534
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/ume/W2PD17514-I4PD23060774.jpg",
+      "label": "ume",
+      "pred": "uchen",
+      "prob_ume": 0.4968777000904083
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/ume/W3CN11633-I3CN116580005.jpg",
+      "label": "ume",
+      "pred": "uchen",
+      "prob_ume": 0.45208024978637695
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/ume/W3CN21413-I2KG2202860536.jpg",
+      "label": "ume",
+      "pred": "ume",
+      "prob_ume": 0.9999990463256836
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/ume/W3CN21798-I4CN123460005.jpg",
+      "label": "ume",
+      "pred": "uchen",
+      "prob_ume": 0.0006142983329482377
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/ume/W3CN4061-I3CN64760512.jpg",
+      "label": "ume",
+      "pred": "ume",
+      "prob_ume": 0.8909825682640076
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/ume/W3CN644-I3CN6460652.jpg",
+      "label": "ume",
+      "pred": "uchen",
+      "prob_ume": 0.032793644815683365
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/ume/W3CN786-I3CN7910005.jpg",
+      "label": "ume",
+      "pred": "ume",
+      "prob_ume": 0.9999990463256836
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/ume/W3CN8231-I3CN82600005.jpg",
+      "label": "ume",
+      "pred": "ume",
+      "prob_ume": 0.9999237060546875
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/ume/W3CN8329-I3CN83570846.jpg",
+      "label": "ume",
+      "pred": "ume",
+      "prob_ume": 0.7273164987564087
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/ume/W3JT13691-I3JT137050066.jpg",
+      "label": "ume",
+      "pred": "uchen",
+      "prob_ume": 0.19064322113990784
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/ume/W3MS155-I1KG227891302.jpg",
+      "label": "ume",
+      "pred": "ume",
+      "prob_ume": 0.9999719858169556
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/ume/W3PD988-I3PD13550824.jpg",
+      "label": "ume",
+      "pred": "ume",
+      "prob_ume": 0.9970397353172302
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/ume/W3PD989-I3PD10890005.jpg",
+      "label": "ume",
+      "pred": "ume",
+      "prob_ume": 0.705629289150238
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/ume/W4CZ58520-I4CZ750920618.jpg",
+      "label": "ume",
+      "pred": "uchen",
+      "prob_ume": 0.028447365388274193
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/ume/W4PD1703-I4PD17140005.jpg",
+      "label": "ume",
+      "pred": "ume",
+      "prob_ume": 0.5697239637374878
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/ume/W8LS16434-I8LS164450005.jpg",
+      "label": "ume",
+      "pred": "ume",
+      "prob_ume": 0.9901788234710693
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/ume/W8LS16555-I8LS165890026.jpg",
+      "label": "ume",
+      "pred": "ume",
+      "prob_ume": 0.9999986886978149
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/ume/W8LS17770-I8LS177950005.jpg",
+      "label": "ume",
+      "pred": "ume",
+      "prob_ume": 0.9999395608901978
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/ume/W8LS19804-I8LS198060061.jpg",
+      "label": "ume",
+      "pred": "ume",
+      "prob_ume": 0.9403765201568604
+    },
+    {
+      "path": "/root/script-classification-model-train/benchmark/benchmark/ume/W8LS20177-I8LS201790005.jpg",
+      "label": "ume",
+      "pred": "ume",
+      "prob_ume": 0.644355058670044
+    }
+  ]
+}

without_preprocess/best_checkpoint.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8809ead826cfe08fbdfa3ca0659c858f2d3e98c21e62d63811905a6dd0c44abc
+size 86674972

without_preprocess/best_checkpoint_name.txt

@@ -0,0 +1 @@
+best_stage_c_last_blocks.pt

without_preprocess/classification_report.txt

@@ -0,0 +1,8 @@
+              precision    recall  f1-score   support
+
+       uchen       0.37      0.98      0.54        99
+         ume       1.00      0.79      0.88       768
+
+    accuracy                           0.81       867
+   macro avg       0.68      0.88      0.71       867
+weighted avg       0.93      0.81      0.84       867

without_preprocess/confusion_matrix.json

@@ -0,0 +1,16 @@
+{
+  "labels": [
+    "uchen",
+    "ume"
+  ],
+  "matrix": [
+    [
+      97,
+      2
+    ],
+    [
+      165,
+      603
+    ]
+  ]
+}

without_preprocess/final_model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:906a64dc6248dd54c08ad71e84cb223dcc22f2e7f613d3157d471908a5c6256f
+size 86672201

without_preprocess/model_card.json

@@ -0,0 +1,15 @@
+{
+  "variant": "without_preprocess",
+  "experiment": "uchen_ume_binary",
+  "best_checkpoint": "best_stage_c_last_blocks.pt",
+  "val_macro_f1": 0.7705722639933166,
+  "val_accuracy": 0.8461538461538461,
+  "epoch": 3,
+  "test_metrics": {
+    "loss": 0.48820294297059763,
+    "accuracy": 0.8073817762399077,
+    "macro_f1": 0.7078823289680483,
+    "weighted_f1": 0.8394339697286689,
+    "auc_roc": 0.9698679503367003
+  }
+}

without_preprocess/results.json

@@ -0,0 +1,17 @@
+{
+  "experiment": "uchen_ume_binary",
+  "stage_run": "test",
+  "test_metrics": {
+    "loss": 0.48820294297059763,
+    "accuracy": 0.8073817762399077,
+    "macro_f1": 0.7078823289680483,
+    "weighted_f1": 0.8394339697286689,
+    "auc_roc": 0.9698679503367003
+  },
+  "history": {},
+  "report": "              precision    recall  f1-score   support\n\n       uchen       0.37      0.98      0.54        99\n         ume       1.00      0.79      0.88       768\n\n    accuracy                           0.81       867\n   macro avg       0.68      0.88      0.71       867\nweighted avg       0.93      0.81      0.84       867\n",
+  "splits_file": "/root/script-classification-model-train/experiments/uchen_ume_binary/checkpoints/uchen_ume_binary/splits.json",
+  "skip_stage_c": false,
+  "stage_c_skip_reason": null,
+  "best_checkpoint": "best_stage_c_last_blocks.pt"
+}