nixos/test-driver: use a variety of different Tesseract settings for OCR

When performing OCR, some of the Tesseract settings perform better than
others on a variety of different workloads, but they mostly take
~negligible incremental time to run compared to the overhead of running
the ImageMagick filters.

After this commit, we try using all three of the current Tesseract
models (classic, LSTM, and classic+LSTM) to generate output text. This
fixes chromium-90's tests at release-20.09, and should make cases where
you're looking for *specific* text better, with the tradeoff of running
Tesseract multiple times.

To make it sensible to cherrypick this into release-20.09, this doesn't
change the existing API surface for the test driver. In particular,
get_screen_text continues to have the existing behaviour.
This commit is contained in:
Luke Granger-Brown 2021-04-23 15:02:36 +00:00
parent 3ac9de55b0
commit 4de343cccf
2 changed files with 70 additions and 33 deletions

View File

@ -186,6 +186,25 @@ start_all()
</para>
</listitem>
</varlistentry>
<varlistentry>
<term>
<methodname>get_screen_text_variants</methodname>
</term>
<listitem>
<para>
Return a list of different interpretations of what is currently visible
on the machine's screen using optical character recognition. The number
and order of the interpretations is not specified and is subject to
change, but if no exception is raised at least one will be returned.
</para>
<note>
<para>
This requires passing <option>enableOCR</option> to the test attribute
set.
</para>
</note>
</listitem>
</varlistentry>
<varlistentry>
<term>
<methodname>get_screen_text</methodname>
@ -350,7 +369,8 @@ start_all()
<para>
Wait until the supplied regular expressions matches the textual contents
of the screen by using optical character recognition (see
<methodname>get_screen_text</methodname>).
<methodname>get_screen_text</methodname> and
<methodname>get_screen_text_variants</methodname>).
</para>
<note>
<para>

View File

@ -1,7 +1,7 @@
#! /somewhere/python3
from contextlib import contextmanager, _GeneratorContextManager
from queue import Queue, Empty
from typing import Tuple, Any, Callable, Dict, Iterator, Optional, List
from typing import Tuple, Any, Callable, Dict, Iterator, Optional, List, Iterable
from xml.sax.saxutils import XMLGenerator
import queue
import io
@ -205,6 +205,37 @@ class Logger:
self.xml.endElement("nest")
def _perform_ocr_on_screenshot(
screenshot_path: str, model_ids: Iterable[int]
) -> List[str]:
if shutil.which("tesseract") is None:
raise Exception("OCR requested but enableOCR is false")
magick_args = (
"-filter Catrom -density 72 -resample 300 "
+ "-contrast -normalize -despeckle -type grayscale "
+ "-sharpen 1 -posterize 3 -negate -gamma 100 "
+ "-blur 1x65535"
)
tess_args = f"-c debug_file=/dev/null --psm 11"
cmd = f"convert {magick_args} {screenshot_path} tiff:{screenshot_path}.tiff"
ret = subprocess.run(cmd, shell=True, capture_output=True)
if ret.returncode != 0:
raise Exception(f"TIFF conversion failed with exit code {ret.returncode}")
model_results = []
for model_id in model_ids:
cmd = f"tesseract {screenshot_path}.tiff - {tess_args} --oem {model_id}"
ret = subprocess.run(cmd, shell=True, capture_output=True)
if ret.returncode != 0:
raise Exception(f"OCR failed with exit code {ret.returncode}")
model_results.append(ret.stdout.decode("utf-8"))
return model_results
class Machine:
def __init__(self, args: Dict[str, Any]) -> None:
if "name" in args:
@ -637,43 +668,29 @@ class Machine:
"""Debugging: Dump the contents of the TTY<n>"""
self.execute("fold -w 80 /dev/vcs{} | systemd-cat".format(tty))
def _get_screen_text_variants(self, model_ids: Iterable[int]) -> List[str]:
with tempfile.TemporaryDirectory() as tmpdir:
screenshot_path = os.path.join(tmpdir, "ppm")
self.send_monitor_command(f"screendump {screenshot_path}")
return _perform_ocr_on_screenshot(screenshot_path, model_ids)
def get_screen_text_variants(self) -> List[str]:
return self._get_screen_text_variants([0, 1, 2])
def get_screen_text(self) -> str:
if shutil.which("tesseract") is None:
raise Exception("get_screen_text used but enableOCR is false")
magick_args = (
"-filter Catrom -density 72 -resample 300 "
+ "-contrast -normalize -despeckle -type grayscale "
+ "-sharpen 1 -posterize 3 -negate -gamma 100 "
+ "-blur 1x65535"
)
tess_args = "-c debug_file=/dev/null --psm 11 --oem 2"
with self.nested("performing optical character recognition"):
with tempfile.NamedTemporaryFile() as tmpin:
self.send_monitor_command("screendump {}".format(tmpin.name))
cmd = "convert {} {} tiff:- | tesseract - - {}".format(
magick_args, tmpin.name, tess_args
)
ret = subprocess.run(cmd, shell=True, capture_output=True)
if ret.returncode != 0:
raise Exception(
"OCR failed with exit code {}".format(ret.returncode)
)
return ret.stdout.decode("utf-8")
return self._get_screen_text_variants([2])[0]
def wait_for_text(self, regex: str) -> None:
def screen_matches(last: bool) -> bool:
text = self.get_screen_text()
matches = re.search(regex, text) is not None
variants = self.get_screen_text_variants()
for text in variants:
if re.search(regex, text) is not None:
return True
if last and not matches:
self.log("Last OCR attempt failed. Text was: {}".format(text))
if last:
self.log("Last OCR attempt failed. Text was: {}".format(variants))
return matches
return False
with self.nested("waiting for {} to appear on screen".format(regex)):
retry(screen_matches)