darktable/lua-scripts-w-api
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

face_recognition.lua script face some stability and performance issues on an i5-6600, 32 GB and GTX 1060 6GB:

Browse Source 
1. Taking to much time – almost a day for 3000 images
    2. Crashing due to lines malformation in facerecognition.txt – some times face_recognition Python script concatenates two output lines in just once, what is not expected by face_recognition.lua script
    3. Crashing the whole system in Linux – I keep receiving “bash: fork: resource temporarily unavailable” trying to process a too big (starting from some hundreds) image list. Some other applications starts to crash at this point. It seems related to the attach tag call because commenting this instruction (for debugging purposes) eliminates the crashes.
After debugging the Lua script and checking Lua and SQL darktable out logs, I decided to make the following changes:
    1. Remove an inner loop in the processing results step – helps preventing issue 1
    2. Remove a loop in the export step – helps preventing issue 1
    3. Remove duplicate tags (when there are more more than a reference face for each person) while facerecognition.txt is read – helps preventing issue 1 and 3
    4. Make a sanity check with processing images, skip any malformation data – helps preventing issue 2
    5. Reduces the sqlite access by caching tags that where already created for a previous image – helps preventing issue 3

Those changes allowed may thousands (more than 30K) of images to be processed in half a day instead of several days.
This commit is contained in:
Piter Dias
2020-04-15 14:03:58 -03:00
parent a582f0229f
commit 9870ddbe86
1 changed files with 73 additions and 67 deletions
Download Patch File Download Diff File Expand all files Collapse all files

140
contrib/face_recognition.lua
View File

@@ -63,7 +63,7 @@ du.check_min_api_version("5.0.0", "face_recognition")
gettext.bindtextdomain("face_recognition", dt.configuration.config_dir.."/lua/locale/")
local function _(msgid)
return gettext.dgettext("face_recognition", msgid)
return gettext.dgettext("face_recognition", msgid)
end
-- preferences
@@ -92,8 +92,10 @@ local function build_image_table(images)
end
for _,img in ipairs(images) do
image_table[img] = tmp_dir .. df.get_basename(img.filename) .. file_extension
cnt = cnt + 1
if img ~= nil then
image_table[tmp_dir .. df.get_basename(img.filename) .. file_extension] = img
cnt = cnt + 1
end
end
return image_table, cnt
@@ -103,17 +105,12 @@ local function stop_job(job)
job.valid = false
end
local function do_export(img_tbl)
local function do_export(img_tbl, images)
local exporter = nil
local upsize = false
local upscale = false
local ff = fc.export_format.value
local height = dt.preferences.read(MODULE, "max_height", "integer")
local width = dt.preferences.read(MODULE, "max_width", "integer")
local images = 0
for k,v in pairs(img_tbl) do
images = images + 1
end
-- get the export format parameters
if string.match(ff, "JPEG") then
@@ -134,7 +131,7 @@ local function do_export(img_tbl)
local exp_cnt = 0
local percent_step = 1.0 / images
job.percent = 0.0
for img,export in pairs(img_tbl) do
for export,img in pairs(img_tbl) do
exp_cnt = exp_cnt + 1
dt.print(string.format(_("Exporting image %i of %i images"), exp_cnt, images))
exporter:write_image(img, export, upsize)
@@ -188,7 +185,7 @@ local function ignoreByTag (image, ignoreTags)
end
end
end
return ignoreImage
end
@@ -221,25 +218,25 @@ local function face_recognition ()
if nrCores < 1 then
nrCores = -1
end
-- Split ignore tags (if any)
ignoreTags = {}
for tag in string.gmatch(ignoreTagString, '([^,]+)') do
table.insert (ignoreTags, tag)
dt.print_log ("Face recognition: Ignore tag: " .. tag)
end
-- list of exported images
local image_table, cnt = build_image_table(dt.gui.action_images)
if cnt > 0 then
local success = do_export(image_table)
local success = do_export(image_table, cnt)
if success then
-- do the face recognition
local img_list = {}
for img,v in pairs(image_table) do
for v,_ in pairs(image_table) do
table.insert (img_list, v)
end
@@ -248,7 +245,7 @@ local function face_recognition ()
dt.print_log ("Face recognition: Path to unknown images: " .. path)
os.setlocale("C")
local tolerance = dt.preferences.read(MODULE, "tolerance", "float")
local command = bin_path .. " --cpus " .. nrCores .. " --tolerance " .. tolerance .. " " .. knownPath .. " " .. path .. " > " .. OUTPUT
os.setlocale()
dt.print_log("Face recognition: Running command: " .. command)
@@ -258,61 +255,77 @@ local function face_recognition ()
-- Open output file
local f = io.open(OUTPUT, "rb")
if not f then
dt.print(_("Face recognition failed"))
else
dt.print(_("Face recognition finished"))
f:close ()
end
-- Read output
dt.print(_("processing results..."))
local result = {}
for line in io.lines(OUTPUT) do
if not string.match(line, "^WARNING:") then
local tags_list = {}
local tag_object = {}
for line in io.lines(OUTPUT) do
if not string.match(line, "^WARNING:") and line ~= "" and line ~= nil then
local file, tag = string.match (line, "(.*),(.*)$")
tag = string.gsub (tag, "%d*$", "")
dt.print_log ("File:"..file .." Tag:".. tag)
if result[file] ~= nil then
table.insert (result[file], tag)
tag_object = {}
if result[file] == nil then
tag_object[tag] = true
result[file] = tag_object
else
result[file] = {tag}
tag_object = result[file]
tag_object[tag] = true
result[file] = tag_object
end
end
end
-- Attach tags
local result_index = 0
for file,tags in pairs(result) do
result_index = result_index +1
-- Find image in table
for img,file2 in pairs(image_table) do
if file == file2 then
for _,t in ipairs (tags) do
-- Check if image is ignored
if ignoreByTag (img, ignoreTags) then
dt.print_log("Face recognition: Ignoring image with ID " .. img.id)
else
-- Check of unrecognized unknown_person
if t == "unknown_person" then
t = unknownTag
end
-- Check of unrecognized no_persons_found
if t == "no_persons_found" then
t = nonpersonsfoundTag
end
if t ~= "" and t ~= nil then
dt.print_log ("ImgId:" .. img.id .. " Tag:".. t)
-- Create tag if it does not exists
local tag = dt.tags.create (t)
img:attach_tag (tag)
img = image_table[file]
if img == nil then
dt.print_log("Face recognition: Ignoring face recognition entry: " .. file)
else
for t,_ in pairs (tags) do
-- Check if image is ignored
if ignoreByTag (img, ignoreTags) then
dt.print_log("Face recognition: Ignoring image with ID " .. img.id)
else
-- Check of unrecognized unknown_person
if t == "unknown_person" then
t = unknownTag
end
-- Check of unrecognized no_persons_found
if t == "no_persons_found" then
t = nonpersonsfoundTag
end
if t ~= "" and t ~= nil then
dt.print_log ("ImgId:" .. img.id .. " Tag:".. t)
-- Create tag if it does not exist
if tags_list[t] == nil then
tag = dt.tags.create (t)
tags_list[t] = tag
else
tag = tags_list[t]
end
img:attach_tag (tag)
end
end
end
end
end
dt.print(_("face recognition complete"))
cleanup(img_list)
dt.print_log("img_list cleaned-up")
dt.print_log("face recognition complete")
dt.print(_("face recognition complete"))
else
dt.print(_("image export failed"))
return
@@ -321,8 +334,6 @@ local function face_recognition ()
dt.print(_("no images selected"))
return
end
end
-- build the interface
@@ -376,7 +387,7 @@ fc.known_image_path = dt.new_widget("file_chooser_button"){
is_directory = true,
changed_callback = function(this)
dt.preferences.write(MODULE, "known_image_path", "directory", this.value)
end
end
}
fc.export_format = dt.new_widget("combobox"){
@@ -403,9 +414,9 @@ fc.height = dt.new_widget("entry"){
fc.execute = dt.new_widget("button"){
label = "detect faces",
clicked_callback = function(this)
clicked_callback = function(this)
face_recognition()
end
end
}
local widgets = {
@@ -422,14 +433,14 @@ local widgets = {
if dt.configuration.running_os == "windows" or dt.configuration.running_os == "macos" then
table.insert(widgets, df.executable_path_widget({"face_recognition"}))
end
table.insert(widgets, dt.new_widget("section_label"){ label = _("processing options")})
table.insert(widgets, fc.tolerance)
table.insert(widgets, fc.num_cores)
table.insert(widgets, fc.export_format)
table.insert(widgets, dt.new_widget("box"){
orientation = "horizontal",
dt.new_widget("label"){ label = _("width ")},
fc.width,
table.insert(widgets, dt.new_widget("section_label"){ label = _("processing options")})
table.insert(widgets, fc.tolerance)
table.insert(widgets, fc.num_cores)
table.insert(widgets, fc.export_format)
table.insert(widgets, dt.new_widget("box"){
orientation = "horizontal",
dt.new_widget("label"){ label = _("width ")},
fc.width,
})
table.insert(widgets, dt.new_widget("box"){
orientation = "horizontal",
@@ -439,18 +450,13 @@ table.insert(widgets, dt.new_widget("box"){
table.insert(widgets, fc.execute)
fc.widget = dt.new_widget("box"){
orientation = vertical,
reset_callback = function(this)
orientation = vertical,
reset_callback = function(this)
reset_preferences()
end,
table.unpack(widgets),
end,
table.unpack(widgets),
}
--fc.tolerance.value = dt.preferences.read(MODULE, "tolerance", "float")
-- Register
--dt.register_storage("module_face_recognition", _("Face recognition"), show_status, face_recognition)
dt.register_lib(
"face_recognition", -- Module name
_("face recognition"), -- Visible name