FFmpeg decoding .mp4 video file

Question

I'm working on a project that needs to open .mp4 file format, read it's frames 1 by 1, decode them and encode them with better type of lossless compression and save them into a file.

Please correct me if i'm wrong with order of doing things, because i'm not 100% sure how this particular thing should be done. From my understanding it should go like this:

1. Open input .mp4 file
2. Find stream info -> find video stream index
3. Copy codec pointer of found video stream index into AVCodecContext type pointer
4. Find decoder -> allocate codec context -> open codec
5. Read frame by frame -> decode the frame -> encode the frame -> save it into a file

So far i encountered couple of problems. For example, if i want to save a frame using av_interleaved_write_frame() function, i can't open input .mp4 file using avformat_open_input() since it's gonna populate filename part of the AVFormatContext structure with input file name and therefore i can't "write" into that file. I've tried different solution using av_guess_format() but when i dump format using dump_format() i get nothing so i can't find stream information about which codec is it using.

So if anyone have any suggestions, i would really appreciate them. Thank you in advance.

Solution

See the "detailed description" in the muxing docs. You:

1. set ctx->oformat using av_guess_format
2. set ctx->pb using avio_open2
3. call avformat_new_stream for each stream in the output file. If you're re-encoding, this is by adding each stream of the input file into the output file.
4. call avformat_write_header
5. call av_interleaved_write_frame in a loop
6. call av_write_trailer
7. close the file (avio_close) and clear up all allocated memory

Comments

[Eugene]

Are you trying to convert a .mp4 to a series of lossless still images?

[Sir DrinksCoffeeALot]

Well i'm trying to split .mp4 file into frames, compress them and send them through a network. On the other end those frames will get concatenated back into a .mp4 file.

[Eugene]

What's the point of that? And compressing lossless images is pointless.The mp4 format is already designed as an efficient container. This process is slow, processor intensive, and bandwidth inefficient, etc.

[Sir DrinksCoffeeALot]

It's because i need to find out differences between multiple codecs, main goal is to split 4k video into frames, compress them and send them through 1Gbit connection. So i need to find out which codec would be the best compression-wise. And as an example i have HD file on which i need to do experiments.

[Sir DrinksCoffeeALot]

I'm not using static build of ffmpeg, i'm using their API in my own project.

[Eugene]

FFmpeg is the command line utility. Which library and language are you using? Libavcodec?

[Sir DrinksCoffeeALot]

I'm using libavcodec/format/device/filter/util/postproc/swscale. I'm writing in C. Basicly it's an old dev build of FFmpeg.

[Gyan]

FFmpeg is a project containing a bunch of libraries for A/V manipulation and an optional set of binaries such as ffmpeg, ffplay..etc which allow you to use the functions available in the libraries.

[Sir DrinksCoffeeALot]

Do you know how can i copy content of AVFrame structure to a char* buffer, so i can send that buffer using winsock2's send() function to a different process? I'm trying to create local server-client communication to measure time needed to send a frame using different compression methods.

[Ronald S. Bultje]

for (y = 0; y < height; y++) memcpy(ptr + y *width, frame->data[0] + y * frame->linesize[0], width). You can also use avcodec_encode_video2() to compress frames (instead of gzip or so, which is what you're probably trying to do).

[Sir DrinksCoffeeALot]

Hmm if i do it like that rather than copying whole AVFrame structure, considering i'm copying raw video frames, would i be able to encode that frame (based only on frame->data[0] and frame->linesize[0]) on the recipient side using avcodec_encode_video2() function?

[Sir DrinksCoffeeALot]

About compression, i was planning to use lzw. I will take closer look into avcodec_encode_video2() compression-wise. Btw, thank you for your responses i really appreciate it.

[Ronald S. Bultje]

You would do the same for data[1-2] with linesize[1-2] (assuming some planar YUV format), sorry, forgot to mention that; or rather, it depends on the number of planes for your pixfmt (see pixdesc).

[Sir DrinksCoffeeALot]

I succeeded sending single decoded frame from one process to another one using send/recv functions. I determined size of char* buffer needed to store a frame of type PIX_FMT_RGB24 using avpicture_get_size(), copying frame->data[0] was done like this : memcpy(buffer + y * frame->linesize[0], frame->data[0] + y*frame->linesize[0], width * 3); and writing to a .ppm file on client side was done using: fwrite(buffer + y * width * 3, 1, width * 3, pf);. Basicly i just needed to multiply width from your code by 3 (since its RGB24) or use exact data stored in frame->linesize[0].

[Sir DrinksCoffeeALot]

I have one more question before i hopefully stop bugging you about FFmpeg, for example if i want to use LZW compression method which is already implemented in FFmpeg, how would i do it using avcodec_encode_video2() function? I guess i need to populate AVCodecContext variable with different parameters before calling the encode funtion?

[Ronald S. Bultje]

ffmpeg has no vanilla lzw encoder. There's a tiff and a gif encoder, which use lzw internally...

[Sir DrinksCoffeeALot]

Yea i saw tiff uses LZW internally, was hoping i could somehow implement their source into mine. Ill try to do that today.