S3CUR commited on
Commit
f08b4ed
·
verified ·
1 Parent(s): 3800bd2

Add hero image block with demo outputs

Browse files
Files changed (1) hide show
  1. README.md +19 -24
README.md CHANGED
@@ -28,6 +28,23 @@ The repo demonstrates that the published dataset is self-sufficient: with
28
  no other inputs you can `pip install -r requirements.txt`, run `train.py`,
29
  and reproduce the same per-slot accuracy reported in the dataset card.
30
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
31
  ## Layout
32
 
33
  ```
@@ -117,30 +134,8 @@ embedded) and picks a clean test-split frame to decode. Writes
117
  | `sample_rollover_d4.jpg` | `06629789` (662,978.9 gal) | d4 = 9 mid-roll (single rollover edge) |
118
  | `sample_rollover_d6d7.jpg` | `06615699` (661,569.9 gal) | d6 = 9 **and** d7 = 9 (double rollover, hardest case) |
119
 
120
- All three should match digit-for-digit. Reference annotated outputs are
121
- checked in next to each input.
122
-
123
- #### Reference outputs
124
-
125
- Each card shows the chosen digit, θ, the consensus confidence, and both
126
- voters (CLA + P90). A grey informational chip means that head's vote
127
- disagreed with the consensus — useful for eyeballing where the heads
128
- diverge.
129
-
130
- <p align="center">
131
- <img src="demo_sample_midsnap_d6.jpg" width="420" alt="sample_midsnap_d6.jpg → 06672429"><br>
132
- <em>sample_midsnap_d6.jpg → <code>06672429</code> (d6 just past park-2)</em>
133
- </p>
134
-
135
- <p align="center">
136
- <img src="demo_sample_rollover_d4.jpg" width="420" alt="sample_rollover_d4.jpg → 06629789"><br>
137
- <em>sample_rollover_d4.jpg → <code>06629789</code> (d4 = 9 mid-roll)</em>
138
- </p>
139
-
140
- <p align="center">
141
- <img src="demo_sample_rollover_d6d7.jpg" width="420" alt="sample_rollover_d6d7.jpg → 06615699"><br>
142
- <em>sample_rollover_d6d7.jpg → <code>06615699</code> (d6 = 9 <strong>and</strong> d7 = 9, double rollover)</em>
143
- </p>
144
 
145
  ## How the per-slot decision is made
146
 
 
28
  no other inputs you can `pip install -r requirements.txt`, run `train.py`,
29
  and reproduce the same per-slot accuracy reported in the dataset card.
30
 
31
+ <p align="center">
32
+ <img src="demo_sample_midsnap_d6.jpg" width="900" alt="annotated reading: 06672429, d6 mid-snap"/><br/>
33
+ <em><code>06672429</code> — d6 at θ=97° (just past park-2, mid-snap stress case)</em>
34
+ </p>
35
+
36
+ <p align="center">
37
+ <img src="demo_sample_rollover_d4.jpg" width="900" alt="annotated reading: 06629789, d4 mid-roll"/><br/>
38
+ <em><code>06629789</code> — d4 = 9 mid-roll (single rollover edge)</em>
39
+ </p>
40
+
41
+ <p align="center">
42
+ <img src="demo_sample_rollover_d6d7.jpg" width="900" alt="annotated reading: 06615699, double rollover"/><br/>
43
+ <em><code>06615699</code> — d6 = 9 <strong>and</strong> d7 = 9 (double rollover, hardest case)</em>
44
+ </p>
45
+
46
+ <p align="center"><sub>Each card shows the chosen digit, θ, the consensus confidence, and both voters (CLA + P90). A grey informational chip means that head disagreed with the consensus.</sub></p>
47
+
48
  ## Layout
49
 
50
  ```
 
134
  | `sample_rollover_d4.jpg` | `06629789` (662,978.9 gal) | d4 = 9 mid-roll (single rollover edge) |
135
  | `sample_rollover_d6d7.jpg` | `06615699` (661,569.9 gal) | d6 = 9 **and** d7 = 9 (double rollover, hardest case) |
136
 
137
+ All three should match digit-for-digit. Annotated reference outputs are
138
+ shown at the top of this card.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
139
 
140
  ## How the per-slot decision is made
141