[FFmpeg-devel] [PATCH v4 4/4] fate/jpegxl_anim: add demuxer fate test for jpegxl_anim

Wed Jun 28 02:46:33 EEST 2023

On 6/27/2023 8:32 PM, Leo Izen wrote:
> On 6/27/23 19:25, James Almer wrote:
>> On 6/26/2023 12:49 PM, Leo Izen wrote:
>>> Adds a fate test for the jpegxl_anim demuxer, that should allow testing
>>> for true positives and false positives for animated jpegxl files. Note
>>> that two of the test cases are not animated, in order to help sort out
>>> false positives.
>>>
>>> Signed-off-by: <leo.izen at gmail.com>
>>> ---
>>>   tests/Makefile                         |  1 +
>>>   tests/fate/jxl.mak                     | 16 ++++++++++++++++
>>>   tests/ref/fate/jxl-anim-demux-belgium  |  6 ++++++
>>>   tests/ref/fate/jxl-anim-demux-icos4d   |  6 ++++++
>>>   tests/ref/fate/jxl-anim-demux-lenna256 |  7 +++++++
>>>   tests/ref/fate/jxl-anim-demux-newton   |  6 ++++++
>>>   6 files changed, 42 insertions(+)
>>>   create mode 100644 tests/fate/jxl.mak
>>>   create mode 100644 tests/ref/fate/jxl-anim-demux-belgium
>>>   create mode 100644 tests/ref/fate/jxl-anim-demux-icos4d
>>>   create mode 100644 tests/ref/fate/jxl-anim-demux-lenna256
>>>   create mode 100644 tests/ref/fate/jxl-anim-demux-newton
>>>
>>> diff --git a/tests/Makefile b/tests/Makefile
>>> index e09f30a0fc..7b80762e81 100644
>>> --- a/tests/Makefile
>>> +++ b/tests/Makefile
>>> @@ -201,6 +201,7 @@ include $(SRC_PATH)/tests/fate/image.mak
>>>   include $(SRC_PATH)/tests/fate/imf.mak
>>>   include $(SRC_PATH)/tests/fate/indeo.mak
>>>   include $(SRC_PATH)/tests/fate/jpeg2000.mak
>>> +include $(SRC_PATH)/tests/fate/jxl.mak
>>>   include $(SRC_PATH)/tests/fate/libavcodec.mak
>>>   include $(SRC_PATH)/tests/fate/libavdevice.mak
>>>   include $(SRC_PATH)/tests/fate/libavformat.mak
>>> diff --git a/tests/fate/jxl.mak b/tests/fate/jxl.mak
>>> new file mode 100644
>>> index 0000000000..057d3be0e1
>>> --- /dev/null
>>> +++ b/tests/fate/jxl.mak
>>> @@ -0,0 +1,16 @@
>>> +# These two are animated JXL files
>>> +FATE_JPEGXL_ANIM_DEMUX += fate-jxl-anim-demux-newton
>>> +fate-jxl-anim-demux-newton: CMD = framecrc -i 
>>> $(TARGET_SAMPLES)/jxl/newton.jxl -c copy
>>> +FATE_JPEGXL_ANIM_DEMUX += fate-jxl-anim-demux-icos4d
>>> +fate-jxl-anim-demux-icos4d: CMD = framecrc -i 
>>> $(TARGET_SAMPLES)/jxl/icos4d.jxl -c copy
>>> +
>>> +# These two are not animated JXL. They are here to check false 
>>> positives.
>>> +FATE_JPEGXL_ANIM_DEMUX += fate-jxl-anim-demux-belgium
>>> +fate-jxl-anim-demux-belgium: CMD = framecrc -i 
>>> $(TARGET_SAMPLES)/jxl/belgium.jxl -c copy
>>> +FATE_JPEGXL_ANIM_DEMUX += fate-jxl-anim-demux-lenna256
>>> +fate-jxl-anim-demux-lenna256: CMD = framecrc -i 
>>> $(TARGET_SAMPLES)/jxl/lenna-256.jxl -c copy
>>> +
>>> +FATE_JPEGXL_ANIM_DEMUX += $(FATE_JPEGXL_ANIM_DEMUX-yes)
>>> +
>>> +FATE_SAMPLES_FFMPEG-$(call FRAMECRC, JPEGXL_ANIM) += 
>>> $(FATE_JPEGXL_ANIM_DEMUX)
>>> +fate-jxl-anim-demux: $(FATE_JPEGXL_ANIM_DEMUX)
>>> diff --git a/tests/ref/fate/jxl-anim-demux-belgium 
>>> b/tests/ref/fate/jxl-anim-demux-belgium
>>> new file mode 100644
>>> index 0000000000..b2fe5035ac
>>> --- /dev/null
>>> +++ b/tests/ref/fate/jxl-anim-demux-belgium
>>> @@ -0,0 +1,6 @@
>>> +#tb 0: 1/25
>>> +#media_type 0: video
>>> +#codec_id 0: jpegxl
>>> +#dimensions 0: 768x512
>>> +#sar 0: 0/1
>>> +0,          0,          0,        1,       32, 0xa2930a20
>>> diff --git a/tests/ref/fate/jxl-anim-demux-icos4d 
>>> b/tests/ref/fate/jxl-anim-demux-icos4d
>>> new file mode 100644
>>> index 0000000000..eff6ff1f1b
>>> --- /dev/null
>>> +++ b/tests/ref/fate/jxl-anim-demux-icos4d
>>> @@ -0,0 +1,6 @@
>>> +#tb 0: 1/1000
>>> +#media_type 0: video
>>> +#codec_id 0: jpegxl
>>> +#dimensions 0: 48x48
>>> +#sar 0: 0/1
>>> +0,          0,          0,        0,    67898, 0x53b6516b
>>> diff --git a/tests/ref/fate/jxl-anim-demux-lenna256 
>>> b/tests/ref/fate/jxl-anim-demux-lenna256
>>> new file mode 100644
>>> index 0000000000..0bd286a451
>>> --- /dev/null
>>> +++ b/tests/ref/fate/jxl-anim-demux-lenna256
>>> @@ -0,0 +1,7 @@
>>> +#tb 0: 1/25
>>> +#media_type 0: video
>>> +#codec_id 0: jpegxl
>>> +#dimensions 0: 256x256
>>> +#sar 0: 0/1
>>> +0,          0,          0,        1,     4096, 0x2409e9e3
>>> +0,          1,          1,        1,     3992, 0x966dbfcb
>>
>> What this tells me is that the parser needs to do bitstream assembly 
>> after all. Image2 should not propagate a single image split in two 
>> packets like this, at the arbitrary limit of 4kb.
>>
>> Since this format seems to have actual delimiters 
>> (FF_JPEGXL_CODESTREAM_SIGNATURE_LE and 
>> FF_JPEGXL_CONTAINER_SIGNATURE_LE) and even buffer bytes with 
>> ff_jpegxl_collect_codestream_header(), you should then do the assembly 
>> in the parser, much like it's done for bmp, jpg and many other image 
>> formats.
>> The anim demuxer can remain as PARSE_HEADERS so it doesn't run the 
>> frame assembly code.
>>
> 
> The codestream signature is 0xFF 0x0A, which can and frequently will 
> occur unescaped in the middle of a valid file. Detecting the end of the 
> file is also a Hard Problem as it requires an entropy decoder, which I 
> believe you NAK'd for being overly complex for a parser in the first 
> iteration of this a few months ago.

Can you link that first iteration?

> 
> If I choose to assemble a bitstream here with the parser, what will end 
> up happening is that the entire sequence of all input frames will be one 
> large AVPacket, as detecting the end of the image is nontrivial. Is this 
> behavior desirable, compared to what image2dec does, which is emit 
> arbitrary 4k-sized packets?

What do you mean by all input frames? In the case of animated jpegxl, 
the anim demuxer would handle it, not image2. image2 should be used to 
read and output single frame images.

In any case, what would happen is that image2 will propagate packets of 
up to 4kb size, which the parser invoked in demux.c will then consume 
and not return anything until it has decided it has a complete coded 
image a decoder can then use to fully decode and output a frame, or a 
muxer encapsulate. Said complete coded image is output as a single 
buffer which is propagated to the caller as a single AVPacket.

> 
> If so, I can change it to assemble a packet, but I just want to make 
> sure that behavior is desirable over what I have here.
> 
> - Leo Izen
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
> 
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".