FanLR commited on
Commit
eb8a007
·
verified ·
1 Parent(s): 52f0700

Align demo positioning and task descriptions with the GroundingPI paper

Browse files
Files changed (3) hide show
  1. app.py +5 -5
  2. docs/ui_studio_r8_lock.json +2 -2
  3. grounding_demo/console.py +17 -17
app.py CHANGED
@@ -307,7 +307,7 @@ with gr.Blocks(
307
  css=CSS,
308
  js=JS,
309
  head=HOST_HEAD,
310
- title="GroundingPI · Visual Studio",
311
  theme=gr.themes.Base(font=["Manrope", "sans-serif"], font_mono=["ui-monospace", "monospace"]),
312
  analytics_enabled=False,
313
  ) as demo:
@@ -315,8 +315,8 @@ with gr.Blocks(
315
  case_state = gr.State(first_case["id"])
316
  gr.HTML(
317
  """<header class="studio-header">
318
- <div><h1>Precise Visual Grounding</h1><p class="header-description">From language to spatial evidence. Explore objects, text, and interaction targets in a unified visual workspace.</p></div>
319
- <span class="header-scope">INTERACTIVE RESEARCH DEMO</span>
320
  </header>
321
  """
322
  )
@@ -327,7 +327,7 @@ with gr.Blocks(
327
  """<div class="control-brand">
328
  <img class="xpeng-logo" src="__XPENG_LOGO__" alt="XPENG" width="180" height="46"/>
329
  <h2>Grounding<span>PI</span></h2>
330
- <p>A visual foundation for spatial understanding.</p>
331
  </div>""".replace("__XPENG_LOGO__", LOGO)
332
  )
333
  with gr.Column(elem_id="control-settings"):
@@ -461,7 +461,7 @@ with gr.Blocks(
461
  with gr.Accordion("Native Token Sequence", open=False):
462
  raw = gr.Textbox(label="RAW GAM SPECIAL TOKENS", lines=4, interactive=False)
463
  gr.HTML(
464
- "<footer class='studio-footer'><span>GroundingPI · Visual Evidence, Precisely Grounded</span><span>Research demonstration <i>Qualitative examples, not benchmark results.</i></span></footer>"
465
  )
466
 
467
  reset_outputs = [
 
307
  css=CSS,
308
  js=JS,
309
  head=HOST_HEAD,
310
+ title="GroundingPI · Perception Studio",
311
  theme=gr.themes.Base(font=["Manrope", "sans-serif"], font_mono=["ui-monospace", "monospace"]),
312
  analytics_enabled=False,
313
  ) as demo:
 
315
  case_state = gr.State(first_case["id"])
316
  gr.HTML(
317
  """<header class="studio-header">
318
+ <div><h1>Perception First. Action Next.</h1><p class="header-description">A grounding foundation for physical intelligence. Explore objects, text, and interaction targets through boxes, points, and semantic labels.</p></div>
319
+ <span class="header-scope">GROUNDINGPI · RESEARCH DEMO</span>
320
  </header>
321
  """
322
  )
 
327
  """<div class="control-brand">
328
  <img class="xpeng-logo" src="__XPENG_LOGO__" alt="XPENG" width="180" height="46"/>
329
  <h2>Grounding<span>PI</span></h2>
330
+ <p>Visual primitives for physical intelligence.</p>
331
  </div>""".replace("__XPENG_LOGO__", LOGO)
332
  )
333
  with gr.Column(elem_id="control-settings"):
 
461
  with gr.Accordion("Native Token Sequence", open=False):
462
  raw = gr.Textbox(label="RAW GAM SPECIAL TOKENS", lines=4, interactive=False)
463
  gr.HTML(
464
+ "<footer class='studio-footer'><span>GroundingPI · Towards Physical Intelligence</span><span>Research demonstration <i>Qualitative examples, not benchmark results.</i></span></footer>"
465
  )
466
 
467
  reset_outputs = [
docs/ui_studio_r8_lock.json CHANGED
@@ -1,6 +1,6 @@
1
  {
2
  "files": {
3
- "app.py": "95f3f0d399588edf789fecdd7099ef4622f180cdb7fe8389989bf0951a89a78f",
4
  "assets/Manrope-OFL.txt": "e01b637272e0cbdfb240184dd98ea5cc671556d9894dae2668d92ab2c906787c",
5
  "assets/Manrope-variable.ttf": "d0639be45d0af36e798172419d7bd173c4bd4f29e2b76cbb69db1d11bf8b0a40",
6
  "assets/SOURCES.md": "3cceb52136b55f86840e9585e3c37faed1adc762452b0fea50aecba696dac7fa",
@@ -38,7 +38,7 @@
38
  "cases/visual_prompt/visual_prompt-02-sensenova_sa1b_visual_2475968.jpg": "ba96beb1cc54682f56e34b1381fefc9b7b95905363cff917bbfdac886e265bac",
39
  "cases/visual_prompt/visual_prompt-03-sensenova_sa1b_visual_1573415.jpg": "2593b4f81c4fbfd70e73df0710fb143af232126daa0a96bc86c72575713475a5",
40
  "grounding_demo/catalog.py": "75e543fed7f0ca018fb41c4b528efc2901aa6a0f05232959297d6d40798f676f",
41
- "grounding_demo/console.py": "dbc1a08c937f18d5c104d4ab18316937d21a2b0d2df9ba2cba296d9ae1e20f16",
42
  "grounding_demo/protocol.py": "aa4e634070ab7a17c6bc477051e97cb02e772e9e37db89d5de9b6497312635b0",
43
  "grounding_demo/render.py": "ba6201d2eda2f3bf7fb0988e35f76303a4c43455d18cc60c4db547d87155d73d"
44
  },
 
1
  {
2
  "files": {
3
+ "app.py": "92ff5e4f5a6737d2283d3462dc2b32c3fcb5048c256bc185559b39bfd5bbd2d0",
4
  "assets/Manrope-OFL.txt": "e01b637272e0cbdfb240184dd98ea5cc671556d9894dae2668d92ab2c906787c",
5
  "assets/Manrope-variable.ttf": "d0639be45d0af36e798172419d7bd173c4bd4f29e2b76cbb69db1d11bf8b0a40",
6
  "assets/SOURCES.md": "3cceb52136b55f86840e9585e3c37faed1adc762452b0fea50aecba696dac7fa",
 
38
  "cases/visual_prompt/visual_prompt-02-sensenova_sa1b_visual_2475968.jpg": "ba96beb1cc54682f56e34b1381fefc9b7b95905363cff917bbfdac886e265bac",
39
  "cases/visual_prompt/visual_prompt-03-sensenova_sa1b_visual_1573415.jpg": "2593b4f81c4fbfd70e73df0710fb143af232126daa0a96bc86c72575713475a5",
40
  "grounding_demo/catalog.py": "75e543fed7f0ca018fb41c4b528efc2901aa6a0f05232959297d6d40798f676f",
41
+ "grounding_demo/console.py": "67a96eaf49ba77b44863969e5e3eb69db9ab862767a4d9a91941439ed7287e37",
42
  "grounding_demo/protocol.py": "aa4e634070ab7a17c6bc477051e97cb02e772e9e37db89d5de9b6497312635b0",
43
  "grounding_demo/render.py": "ba6201d2eda2f3bf7fb0988e35f76303a4c43455d18cc60c4db547d87155d73d"
44
  },
grounding_demo/console.py CHANGED
@@ -23,55 +23,55 @@ CATEGORY_ROUTES = {"grounding", "dense", "layout"}
23
  TASK_HELP = {
24
  "grounding": [
25
  "Open-Vocabulary Grounding",
26
- "Localize visual instances from category names, with spatial evidence for each prediction.",
27
  "Separate categories with commas. Select a result to inspect its original pixels.",
28
  "car, person"
29
  ],
30
  "dense": [
31
- "Dense Instance Grounding",
32
- "Resolve multiple instances in crowded scenes, from small objects to partially occluded targets.",
33
  "Enter object categories. Inspect overlapping predictions one instance at a time.",
34
  "person, chair"
35
  ],
36
  "referring": [
37
- "Language-Guided Referring",
38
- "Connect a natural-language description to the object it identifies.",
39
  "Describe appearance, position, or relationships in a complete phrase.",
40
  "The person standing beside the table"
41
  ],
42
  "dense_point": [
43
- "Dense Point Localization",
44
- "Represent multiple visual targets as spatial points.",
45
  "Describe the targets to locate. Points indicate positions, not object boundaries.",
46
  "All visible people"
47
  ],
48
  "refer_point": [
49
- "Relational & Affordance Pointing",
50
- "Ground relational descriptions and actionable object parts in image coordinates.",
51
  "Describe an object, relationship, or interaction region. This demo predicts locations only.",
52
  "The handle of the cup"
53
  ],
54
  "visual_prompt": [
55
- "Exemplar-Guided Grounding",
56
- "Use spatial examples to identify visually related instances.",
57
  "Green INPUT boxes come from your prompt, not model predictions. Coordinates use the 0–999 range.",
58
  "Given reference boxes <|box_start|><100><100><200><200><|box_end|> ..."
59
  ],
60
  "ocr": [
61
- "Text as Spatial Evidence",
62
- "Read visible text and associate each transcription with its image region.",
63
  "Use an OCR instruction. Select a transcription to inspect the original pixels.",
64
  "Read and locate the text in the image"
65
  ],
66
  "layout": [
67
- "Document Layout Understanding",
68
- "Map document structure into localized semantic regions.",
69
  "Separate region categories with commas. Regions do not imply a predicted reading order.",
70
  "Title, Text, Table"
71
  ],
72
  "gui": [
73
- "Interface Grounding",
74
- "Translate an interaction instruction into a precise interface location.",
75
  "Describe the element to select. The demo predicts a point; it does not execute clicks.",
76
  "Click the search button"
77
  ]
 
23
  TASK_HELP = {
24
  "grounding": [
25
  "Open-Vocabulary Grounding",
26
+ "Turn category names into visual primitives: what is present, and where it is.",
27
  "Separate categories with commas. Select a result to inspect its original pixels.",
28
  "car, person"
29
  ],
30
  "dense": [
31
+ "Grounding in Dense Scenes",
32
+ "Explore object-level perception in crowded scenes, including small and overlapping instances.",
33
  "Enter object categories. Inspect overlapping predictions one instance at a time.",
34
  "person, chair"
35
  ],
36
  "referring": [
37
+ "Language, Grounded in Context",
38
+ "Link descriptions of appearance, position, and relationships to a specific visual target.",
39
  "Describe appearance, position, or relationships in a complete phrase.",
40
  "The person standing beside the table"
41
  ],
42
  "dense_point": [
43
+ "Dense Object Pointing",
44
+ "Represent multiple objects as localized points: a compact form of visual grounding.",
45
  "Describe the targets to locate. Points indicate positions, not object boundaries.",
46
  "All visible people"
47
  ],
48
  "refer_point": [
49
+ "Interaction-Centered Pointing",
50
+ "Locate object parts and interaction regions from language, connecting perception with action-relevant targets.",
51
  "Describe an object, relationship, or interaction region. This demo predicts locations only.",
52
  "The handle of the cup"
53
  ],
54
  "visual_prompt": [
55
+ "Grounding by Visual Example",
56
+ "Use spatial exemplars to identify related instances through a shared grounding interface.",
57
  "Green INPUT boxes come from your prompt, not model predictions. Coordinates use the 0–999 range.",
58
  "Given reference boxes <|box_start|><100><100><200><200><|box_end|> ..."
59
  ],
60
  "ocr": [
61
+ "Reading with Spatial Context",
62
+ "Bind recognized text to its image region, connecting what is read with where it appears.",
63
  "Use an OCR instruction. Select a transcription to inspect the original pixels.",
64
  "Read and locate the text in the image"
65
  ],
66
  "layout": [
67
+ "Structure, Grounded on the Page",
68
+ "Express document structure as named visual regions, from titles and text blocks to tables.",
69
  "Separate region categories with commas. Regions do not imply a predicted reading order.",
70
  "Title, Text, Table"
71
  ],
72
  "gui": [
73
+ "Grounding Intent in Interfaces",
74
+ "Map an interaction instruction to a visual target within an interface.",
75
  "Describe the element to select. The demo predicts a point; it does not execute clicks.",
76
  "Click the search button"
77
  ]