Invalid bounding polygon computed exception during segmentation

[fattynoparents]

I'm trying to segment a bunch of images, however the segment command fails at one of the images with the following error message:

ERROR    Failed processing /home/user/images/189097.jpg: Invalid bounding polygon computed: 
Self-intersection[224.964516129032 302.148387096774]

What could cause this?
Thanks in advance.

Traceback:

╭─────────────────────────────── Traceback (most recent call last) ────────────────────────────────╮
│ /path/to/kraken/bin/kraken:8 in <module>                                                    │
│                                                                                                  │
│   5 from kraken.kraken import cli                                                                │
│   6 if __name__ == '__main__':                                                                   │
│   7 │   sys.argv[0] = re.sub(r'(-script\.pyw|\.exe)?$', '', sys.argv[0])                         │
│ ❱ 8 │   sys.exit(cli())                                                                          │
│   9                                                                                              │
│                                                                                                  │
│ /path/to/kraken/lib/python3.11/site-packages/click/core.py:1157 in __call__                 │
│                                                                                                  │
│ /path/to/kraken/lib/python3.11/site-packages/click/core.py:1078 in main                     │
│                                                                                                  │
│ /path/to/kraken/lib/python3.11/site-packages/click/core.py:1720 in invoke                   │
│                                                                                                  │
│ /path/to/kraken/lib/python3.11/site-packages/click/core.py:1657 in _process_result          │
│                                                                                                  │
│ /path/to/kraken/lib/python3.11/site-packages/click/core.py:783 in invoke                    │
│                                                                                                  │
│ /path/to/kraken/lib/python3.11/site-packages/kraken/kraken.py:431 in process_pipeline       │
│                                                                                                  │
│   428 │   │   │   │   if len(fc) - 2 == idx:                                                     │
│   429 │   │   │   │   │   ctx.meta['last_process'] = True                                        │
│   430 │   │   │   │   with threadpool_limits(limits=ctx.meta['threads']):                        │
│ ❱ 431 │   │   │   │   │   task(input=input, output=output)                                       │
│   432 │   │   except Exception as e:                                                             │
│   433 │   │   │   logger.error(f'Failed processing {io_pair[0]}: {str(e)}')                      │
│   434 │   │   │   if ctx.meta['raise_failed']:                                                   │
│                                                                                                  │
│ /path/to/kraken/lib/python3.11/site-packages/kraken/kraken.py:154 in segmenter              │
│                                                                                                  │
│   151 │   │   │   │   │   │   │   │     pad=pad,                                                 │
│   152 │   │   │   │   │   │   │   │     mask=mask)                                               │
│   153 │   │   else:                                                                              │
│ ❱ 154 │   │   │   res = blla.segment(im, text_direction, mask=mask, model=models, device=devic   │
│   155 │   │   │   │   │   │   │      raise_on_error=ctx.meta['raise_failed'], autocast=ctx.met   │
│   156 │   except Exception:                                                                      │
│   157 │   │   if ctx.meta['raise_failed']:                                                       │
│                                                                                                  │
│ /path/to/kraken/lib/python3.11/site-packages/kraken/blla.py:342 in segment                  │
│                                                                                                  │
│   339 │   │   suppl_obj = scale_regions([x.boundary for x in suppl_obj], 1/rets['scale'])        │
│   340 │   │   line_regs = scale_regions([x.boundary for x in line_regs], 1/rets['scale'])        │
│   341 │   │                                                                                      │
│ ❱ 342 │   │   _lines = vec_lines(**rets,                                                         │
│   343 │   │   │   │   │   │      regions=line_regs,                                              │
│   344 │   │   │   │   │   │      text_direction=text_direction,                                  │
│   345 │   │   │   │   │   │      suppl_obj=suppl_obj,                                            │
│                                                                                                  │
│ /path/to/kraken/lib/python3.11/site-packages/kraken/blla.py:229 in vec_lines                │
│                                                                                                  │
│   226 │   │   for reg_idx, reg_pol in enumerate(reg_pols):                                       │
│   227 │   │   │   if is_in_region(bl_ls, reg_pol):                                               │
│   228 │   │   │   │   suppl_obj.append(regions[reg_idx])                                         │
│ ❱ 229 │   │   pol = calculate_polygonal_environment(baselines=[bl[1]],                           │
│   230 │   │   │   │   │   │   │   │   │   │   │     im_feats=im_feats,                           │
│   231 │   │   │   │   │   │   │   │   │   │   │     suppl_obj=suppl_obj,                         │
│   232 │   │   │   │   │   │   │   │   │   │   │     topline=topline,                             │
│                                                                                                  │
│ /path/to/kraken/lib/python3.11/site-packages/kraken/lib/segmentation.py:770 in              │
│ calculate_polygonal_environment                                                                  │
│                                                                                                  │
│    767 │   │   │                                                                                 │
│    768 │   │   │   env_up, env_bottom = _calc_roi(line, bounds, baselines[:idx] + baselines[idx  │
│    769 │   │   │                                                                                 │
│ ❱  770 │   │   │   polygons.append(_extract_patch(env_up,                                        │
│    771 │   │   │   │   │   │   │   │   │   │      env_bottom,                                    │
│    772 │   │   │   │   │   │   │   │   │   │      line.astype('int'),                            │
│    773 │   │   │   │   │   │   │   │   │   │      offset_line.astype('int'),                     │
│                                                                                                  │
│ /path/to/kraken/lib/python3.11/site-packages/kraken/lib/segmentation.py:620 in              │
│ _extract_patch                                                                                   │
│                                                                                                  │
│    617 │   │   polygon = np.concatenate(([end_points[-1]], upper_seam, [end_points[0]], bottom_  │
│    618 │   │   polygon = geom.Polygon(polygon)                                                   │
│    619 │   if not polygon.is_valid:                                                              │
│ ❱  620 │   │   raise Exception(f'Invalid bounding polygon computed: {explain_validity(polygon)}  │
│    621 │   polygon = np.array(roi_polygon.intersection(polygon).boundary.coords, dtype=int)      │
│    622 │   return polygon                                                                        │
│    623

[mittagessen]

Does it actually fail or does it just print this message and ignore the line? The latter is expected, the former not.

What could cause this?

Usually polygonization fails for broken lines, e.g. short line fragments, very high curvature, things like that.

[fattynoparents]

Does it actually fail or does it just print this message and ignore the line?

Yes it fails and stops processing other images.

Usually polygonization fails for broken lines, e.g. short line fragments, very high curvature, things like that.

That's weird because I have much worse images that kraken processes fine. If I attach the image, will it help you to understand why this could happen?

EDIT: Now that I think a bit more about it, I remember that the previous dev version worked fine, that is it skipped unsuccessful images and continued the work on other images.

[fattynoparents]

Hi, did you manage to reproduce the error? Do you think it should be easy to fix?

[fattynoparents]

Getting a new exception now, and the segment command stops executing:

TopologyException: side location conflict at 1345.5620723362658 1028.5092864125122. This can occur if the input geometry is invalid.
[08/15/24 11:58:28] ERROR    Failed processing /images/29022.jpg: No Shapely geometry can be created from   kraken.py:433
                             null value

[fattynoparents]

Any news on fixing this?

[mittagessen]

Sorry, it's vacation time in France so as a good assimilating foreigner I've been off all of August.

Could you give me the image it happens on? And try the latest main branch as a bug in the polygonizer got fixed that caused weird bounding polygons (which are prone to triggering these kinds of shapely errors).

[fattynoparents]

Oh, sorry that I got on you, hope you had a good rest :)

I use the latest dev version (I guess?) 5.2.9.dev6, get the exception:

TopologyException: side location conflict at 1345.5620723362658 1028.5092864125122. This can occur if the input geometry is invalid.
[08/28/24 08:14:39] ERROR    Failed processing /images/29022.jpg: No Shapely geometry can be  created from null value

on this image:

[fattynoparents]

Hi, just checking if you had time to find out what can cause the TopologyException during the segmentation?