[Libav-user] AV_CODEC_TYPE_TIFF packet size problems

[wm4]

On Fri, 14 Nov 2014 11:41:54 -0000
coordz at megaroms.co.uk wrote:

> I'm using the AV_CODEC_TYPE_TIFF to work with a 5184x3456 pixel image with
> R16G16B16, i.e. AV_PIX_FMT_RGB48. When I read the image in I get a packet
> of a size of 107665156 bytes which tallies nicely with the 107495424 bytes
> of uncompressed pixel data that should be in the packet. I then use
> sws_scale() to add an alpha channel, i.e. go to AV_PIX_FMT_RGBA64.
> 
> When I write the image data out as AV_PIX_FMT_RGB48 (using sws_scale() to
> remove previously added alpha channel), AV_CODEC_TYPE_TIFF generates a
> packet of 108363208 bytes which, again, seems fine.
> 
> Now I try to write the file out as AV_PIX_FMT_RGBA64, I get an error
> 
> "[tiff @ 000000000F49A0A0] Invalid minimum required packet size 2293265920
> (max allowed is 2147483631)"
> 
> My first worry with this is that for RGBA64 I should get a packet size of
> about 143327232 bytes (the pixel data size) not 2293265920 bytes (about
> 5184 x 3456 x 3 channels x 2 bytes x 2 (unknown factor???)). What's going
> on here? I don't believe this is correct so is this a bug with
> AV_CODEC_TYPE_TIFF?
> 
> Also, looking up the code that generates this error (ff_alloc_packet2() in
> utils.c) it looks like I'm restricted to images that fit in a packet of a
> bit less than MAX_INT which is about a 30MPixel image. I'm assuming this
> is a very difficult restriction to get around in the avcodec code?

Didn't read most of your post; but yes, there are such restrictions.
The code uses "int" instead of size_t in too many places. It seems
you're hitting such a limit.

Fixing this would probably be very hard, but you can try to discuss
this on ffmpeg-devel.

> To work around these issues I'm thinking of using the lossless compression
> in TIFF to try to reduce the packet size. How do I instruct
> AV_CODEC_TYPE_TIFF to do this compression?
> 
> Thanks!