Hello! Thanks for your work. Would you like to share more information about the setting information in ABR simulation?
For example, could you make the SynthVideo and SynthTrace dataset be open-source? Or could you provide a more detailed instruction on how to prepare our own unseen data in Table 3 and to reproduce Figure 12?
As mentioned in code, we can calculate the reward as following:
reward = VIDEO_BIT_RATE[bit_rate] / M_IN_K
- REBUF_PENALTY * rebuf
- SMOOTH_PENALTY * abs(VIDEO_BIT_RATE[bit_rate] - VIDEO_BIT_RATE[last_bit_rate]) / M_IN_K
This calculation method is often heavily influenced by network quality. For example, in environments with higher bandwidth, a higher bitrate can be chosen, resulting in clearer videos and thus a larger reward. However, this does not necessarily indicate that the model performs better across different portions of the network dataset. Is my understanding correct?
Thank you!
Hello! Thanks for your work. Would you like to share more information about the setting information in ABR simulation?
For example, could you make the SynthVideo and SynthTrace dataset be open-source? Or could you provide a more detailed instruction on how to prepare our own unseen data in Table 3 and to reproduce Figure 12?
As mentioned in code, we can calculate the reward as following:
reward = VIDEO_BIT_RATE[bit_rate] / M_IN_K
- REBUF_PENALTY * rebuf
- SMOOTH_PENALTY * abs(VIDEO_BIT_RATE[bit_rate] - VIDEO_BIT_RATE[last_bit_rate]) / M_IN_K
This calculation method is often heavily influenced by network quality. For example, in environments with higher bandwidth, a higher bitrate can be chosen, resulting in clearer videos and thus a larger reward. However, this does not necessarily indicate that the model performs better across different portions of the network dataset. Is my understanding correct?
Thank you!