ApsaraVideo Media Processing:Terms

Job

A job is the abstraction that MPS uses to represent one media processing request. Every job carries three key pieces of information: the input, the output, and the processing parameters.

• Common job types — MPS provides media information analysis jobs, transcoding jobs, transcoding query jobs, snapshot jobs, and video AI jobs.

  • Transcoding jobs include regular transcoding, Narrowband HD™ 1.0 transcoding, Narrowband HD™ 2.0 transcoding, and audio and video enhancement jobs.

  • Video AI jobs include intelligent production, smart tagging, content moderation, media fingerprinting, and copyright watermarking jobs.

• Job submission

  • Submission method — You can submit a job in the MPS console, by calling the software development kits (SDKs) or API operations for the job type, or by triggering a workflow that you configured.

  • Required parameters — For example, when you submit a transcoding job, the core parameters that you must specify include the input file, output path, transcoding template (or transcoding parameters), MPS queue, and priority.

  The time required for running different types of jobs varies. Some jobs are complete immediately after they are submitted. Most jobs, however, require files to be downloaded, processed, and analyzed, which cannot be completed in real time. Therefore, jobs are divided into synchronous and asynchronous invocation modes. The following table compares the two modes.

To retrieve the result of an asynchronous job, use one of the following methods:

• Periodic polling — Each job is identified by a unique JobId, which is returned to the caller when the job is submitted. You can also view the JobId in the job list in the console. Record the JobId and periodically poll the job query operation to retrieve the task result.

• MNS configuration — Configure MNS notifications for a pipeline or workflow to receive the task result in a timely manner. A notification contains the job ID (JobId), the user data (UserData), and the detailed result.

Pipeline

A pipeline is a queue for asynchronous jobs. When you submit an asynchronous job, the job is queued in the pipeline. Within the concurrency limit, queued jobs are scheduled and run in order based on their priority and submission time. Different types of pipelines vary in the number of pipelines that you can create, the features that they support, and their concurrency capabilities. For more information, see Pipeline overview.

Workflow

A workflow is a preset sequence of jobs. After a file is uploaded to the specified Object Storage Service (OSS) bucket, the jobs run in order based on the steps and conditions that you arranged in advance.

Transcoding template

A transcoding template is a collection of processing parameters that is identified by a unique ID. Use a transcoding template to simplify the operations when you create a transcoding job or use a workflow. Based on the creation source, transcoding templates are classified into the following three types. For more information, see Transcoding template overview.

• Custom template — A transcoding template that you create in the MPS console or by calling an API operation. No parameters are customized at the backend.

• Customized template — A transcoding template for which MPS configures customized parameters at the backend based on your personalized requirements. The customized parameters determine the processing result, and the parameters that you configure do not take effect. The customized parameters are not visible to you and cannot be modified on your side.

• Preset template — A transcoding template that MPS provides for different resolutions to fit a specific range of network bandwidths. Preset templates include static preset templates and intelligent preset templates. Static preset templates support regular transcoding, audio transcoding, container format conversion, Narrowband HD™ 1.0, and Narrowband HD™ 2.0. For more information, go to the MPS console or see Preset template details.

Template analysis job

A template analysis job returns the list of preset templates that are available for an input file. Because input files vary in resolution, bitrate, and other attributes, not all preset templates are suitable for a given input file. Therefore, before you use a preset template, call the Submit a template analysis job operation. To retrieve the result, call the QueryAnalysisJobList operation.

Watermark template

A watermark template is a collection of parameters that are used to add a watermark to a video, such as the position, offset, and size of the watermark. Each watermark template is identified by a unique ID. To add a watermark to an output video, specify a watermark template or pass in the corresponding parameters directly.

Transcoding

Transcoding is the process of converting a compressed and encoded audio or video stream into another audio or video stream to suit different network bandwidths, terminal processing capabilities, and user requirements. Transcoding is essentially a decoding process followed by an encoding process. Therefore, the streams before and after conversion may or may not use the same video encoding format. H.264, H.265, and AV1 are the commonly used encoding formats.

Container format conversion

Container format conversion converts the container format of a video or audio file, such as converting an AVI video to MP4. No audio or video encoding or decoding is performed. Instead, the process extracts the compressed video and audio streams from a file in one container format and finishes by packaging them into a file in another container format. Compared with transcoding, container format conversion has the following two characteristics:

• Extremely fast processing — Audio and video encoding and decoding are highly complex and account for most of the transcoding time. Container format conversion requires no encoding or decoding, which saves a large amount of processing time.

• No loss of audio or video quality — Because no decoding (decompression) or encoding (compression) is performed, no compression loss occurs. The converted file is almost identical to the original file in resolution and bitrate.

Resolution

Resolution describes the ability of a video to resolve detail. It is usually expressed as the number of pixels in each dimension. For example, 1280 × 720 indicates that the video is 1280 pixels wide and 720 pixels high. A video with a higher resolution contains more pixels, so the details in the image are finer and the image is clearer.

Resolution is the main factor that determines the bitrate, and different resolutions require different bitrates. A higher resolution generally requires a higher bitrate, but each resolution has its own reasonable bitrate range. Below that range, the image quality is poor. Above that range, the image quality improves little or not at all, and network traffic and storage space are wasted.

Bitrate

Bitrate is the data traffic that a video file uses per unit of time. It is also called the stream or the stream rate, and it is the most important factor for image quality control in video encoding. Bitrate is measured in bits per second (bit/s or bps) and is commonly expressed in Kbps (thousands of bits per second) or Mbps. At the same resolution, a higher bitrate lowers the compression ratio and improves image quality. A higher bitrate also means a higher sampling rate per unit of time and higher data stream accuracy, so the processed file stays closer to the original file and the image is clearer. In exchange, the playback device requires higher decoding capabilities.

A higher bitrate also produces a larger file. The file size is calculated based on the following formula: File size = Duration × Bitrate/8. For example, a 60-minute 720p video file with a bitrate of 1 Mbps, which is common online, is about 3600 seconds × 1 Mb/8 = 450 MB in size.

Frame rate

Frame rate is the unit of measurement for the number of video frames displayed per unit of time, that is, the number of images refreshed per second. It is measured in frames per second (FPS) or hertz (Hz).

A higher frame rate delivers smoother and more lifelike images. In general, 25 to 30 FPS is acceptable, and increasing the frame rate to 60 FPS noticeably improves interactivity and realism. Beyond 75 FPS, however, further improvements in smoothness are difficult to perceive. If the frame rate exceeds the refresh rate of the screen, the display cannot refresh that quickly, so the extra frames and the graphics processing capacity spent on them are wasted. At the same resolution, a higher frame rate requires higher processing capabilities from the graphics card.

Video keyframe interval (GOP)

A group of pictures (GOP) is a group of consecutive images within an MPEG-encoded video or video stream. A GOP starts with an I-frame and ends at the next I-frame. A GOP contains the following image types:

• I-frame (intra coded picture) — Also called an intra-coded frame or keyframe. An I-frame is an independent frame that carries all of its own information and can be decoded without referencing other images. You can think of it as a still image. The first frame in a video sequence is always an I-frame, and each GOP starts with an I-frame.

• P-frame (predictive coded picture) — Also called an inter-frame predictive coded frame. A P-frame must reference the preceding I-frame to be encoded. It represents the difference between the current frame and the previous frame, which can be an I-frame or a P-frame. During decoding, the difference defined by the current frame is superimposed on the previously cached image to generate the final image. Compared with an I-frame, a P-frame usually occupies fewer data bits. However, because a P-frame has complex dependencies on the preceding P-frames and I-frames, it is highly sensitive to transmission errors.

• B-frame (bidirectionally predictive coded picture) — Also called a bidirectionally predictive coded frame. A B-frame records the difference between the current frame and both the preceding and following frames. To decode a B-frame, you must obtain the previously cached image and also decode the following image, and then superimpose both on the data of the current frame to obtain the final image. B-frames provide a high compression ratio but require higher decoding performance.

  The GOP value indicates the keyframe interval, that is, the number of frames between two keyframes, the distance between two IDR frames, and the maximum number of frames in a frame group. In general, at least one keyframe is required for each second of video. Increasing the number of keyframes improves video quality but also increases bandwidth consumption and network load. To obtain the time interval, divide the GOP value in frames by the frame rate. For example, a GOP value of 250 frames at a frame rate of 25 FPS gives a time interval of 10 seconds.

Keep the GOP value within a reasonable range to balance video quality, file size (network bandwidth), and seek behavior (the response speed of dragging and fast-forwarding):

• Increasing the GOP value helps reduce the video file size. However, do not set it too large. An excessively large GOP value distorts the images in the later frames of the GOP and degrades video quality.

• The GOP value is also a key factor that affects the seek response speed. During a seek, the player locates the nearest preceding keyframe of the specified position. A larger GOP value means that the keyframe may be farther from the specified position, so more predictive frames must be decoded and the seek response time (buffering time) is longer.

• Because P-frames and B-frames are more complex than I-frames, an excessively large GOP value produces too many P-frames and B-frames, which reduces encoding efficiency.

• However, if the GOP value is too small, you must increase the output bitrate of the video to keep the image quality from degrading, which increases bandwidth consumption.

Audio and video encoding profile

An encoding profile is a set of specific encoding capabilities for a specific class of applications.

H.264 defines the following three main profiles:

• Baseline — Basic image quality, suitable for mobile devices. This profile supports I-frames and P-frames, and supports only progressive scanning and context-adaptive variable-length coding (CAVLC).

• Main — Mainstream image quality, suitable for standard-definition devices, such as MP4 players with relatively low decoding capabilities, portable video players, PSPs, and iPods. This profile supports I-frames, P-frames, and B-frames, supports both progressive and interlaced scanning, and supports both CAVLC and context-adaptive binary arithmetic coding (CABAC).

• High — High image quality, suitable for devices with high resolution and large screens, such as broadcasting, video disc storage (Blu-ray Discs), and high-definition television. In addition to the capabilities of the Main profile, this profile adds 8 × 8 intra-prediction, custom quantization, lossless video coding, and more YUV formats.

  The main AAC profiles are as follows:

• aac_low: Low Complexity AAC (LC)

• aac_he: High Efficiency AAC (HE-AAC)

• aac_he_v2: High Efficiency AAC version 2 (HE-AACv2)

• aac_ld: Low Delay AAC (LD)

• aac_eld: Enhanced Low Delay AAC (ELD)

Bitrate control and encoding passes

Bitrate control is the process of determining the output bitrate during video encoding. Two settings work together: the bitrate control method, which determines how the output bitrate is allocated, and the number of encoding passes, which determines how many times the encoder processes the video. The following bitrate control methods are commonly used:

• CBR (constant bitrate) — A fixed-bitrate mode in which the file uses one bitrate from beginning to end. Compared with VBR and ABR, CBR produces larger files, and the video quality is not noticeably better than that of VBR and ABR.

• VBR (variable bitrate) — A dynamic-bitrate mode, that is, a non-fixed bitrate. During audio and video encoding, the bitrate is determined on the fly based on the complexity of the input file: higher bitrates are allocated to complex content and lower bitrates to simple content. VBR is usually used with the Two-Pass encoding mode. VBR is suitable for storage scenarios because it makes more reasonable use of limited storage space, but you cannot predict the size of the output file or its bitrate fluctuations.

• ABR (average bitrate) — The default bitrate control method of Alibaba Cloud. ABR is an average-bitrate mode that is an interpolated variant of VBR. LAME created this encoding mode to address the poor size-to-quality ratio of CBR and the unpredictable file size of VBR. Within a specified file size, ABR divides the stream into segments of 50 frames (30 frames is about one second), and it uses relatively low bitrates for low-frequency and less perceptible frequencies and high bitrates for high-frequency content and large dynamic ranges. ABR reaches the specified bitrate within a certain time range: local bitrate peaks can exceed the specified bitrate, but the average bitrate remains constant within a reasonable range and encoding stays dynamic based on complexity. ABR can therefore be regarded as a compromise between VBR and CBR.

• VBV (Video Buffering Verifier) — The video buffering verifier provides a way to keep the bitrate below a maximum value. To use this method, set the peak output bitrate (maxrate) and the buffer size (bufsize). VBV can be used with Two-Pass or CRF encoding. The latter combination is also known as Capped CRF.

  • Bufsize — The size of the video buffer. Set it based on the bitrate fluctuation that you expect. Usually, set bufsize to twice the value of maxrate. If the client has a small cache, set bufsize equal to maxrate. To limit the bitrate, set bufsize to half the value of maxrate or less.

• CRF (constant rate factor) — A quality control factor mode. Video quality is quantified into levels, where 0 is lossless and 51 is the worst. Setting the CRF keeps the overall subjective video quality constant, so the bitrate fluctuates with scene complexity. (Recommended) If you are unsure which CRF to use, set a value between 23 and 29. Adjust it based on image complexity: each increase of 6 halves the bitrate, and each decrease of 6 doubles the bitrate. At the same definition, animation can usually be set higher than live-action footage. CRF delivers better video quality, but you cannot predict the size of the output file or its bitrate fluctuations.

• Capped CRF — The bitrate output by CRF is not fixed. You can use CRF together with VBV to limit the range of bitrate fluctuation and avoid bitrate spikes.

  MPS supports the following encoding passes:

• One-Pass — The default encoding method of Alibaba Cloud. One-Pass encodes faster than Two-Pass.

• Two-Pass — The encoder runs twice to allocate the bitrate precisely and produce a smaller file with higher quality. The first pass analyzes the video and generates a log file, and the second pass encodes based on the analysis results to achieve the best encoding quality. Two-Pass takes more time than One-Pass, so it cannot be used in scenarios that require high transcoding timeliness, such as live streaming and real-time communication. Two-Pass is also unsuitable when the input video is already highly compressed, because blocking artifacts occur.

Region

A region is an Alibaba Cloud service node. Alibaba Cloud services are available in different regions, so you can select services in the region closest to you for lower access latency and a better user experience.

Object Storage Service (OSS)

OSS refers to Alibaba Cloud Object Storage Service. ApsaraVideo Media Processing transcodes the media files that you store in OSS, and the transcoded output files are also stored in OSS. For more information about OSS concepts, see OSS terms.

Bucket

A bucket is a container that you use to store objects. Every object belongs to a bucket. A bucket has various configuration attributes, including the region, access permissions, and storage class. You can create different types of buckets to store different data based on your requirements. For more information, see Bucket in OSS terms.

Object

Objects are the basic unit of data storage in OSS and are also called OSS files. Unlike a traditional file system, OSS has no hierarchical directory structure for objects. An object consists of object metadata (Object Meta), user data (Data), and a file name (Key), and is identified by a Key that is unique within the bucket. Object metadata is a set of key-value pairs that describe attributes of the object, such as the last modified time and the size. You can also store custom information in the metadata. For more information, see Object in OSS terms.

AccessKey pair

An AccessKey, or AK for short, refers to the AccessKey ID and AccessKey secret that are used for access identity verification. OSS verifies the identity of the sender of a request by checking a signature that is generated from the AccessKey ID and the AccessKey secret through symmetric encryption. The AccessKey ID identifies the user. The AccessKey secret is the key that you use to encrypt signature strings and that OSS uses to verify signature strings. Keep the AccessKey secret confidential.

There are three types of AccessKeys:

• An AccessKey that the bucket owner applies for.

• An AccessKey that the bucket owner grants to a third-party requester by using RAM authorization.

• An AccessKey that the bucket owner grants to a third-party requester by using STS.

  For more information, see Create an AccessKey pair.