typo in the abstract? "MobileNetV2-Small is 4.6% more accurate while reducing latency by 5% compared to MobileNetV2". Is the MobileNetV2-Small meant to be MobileNetV3-Small? And I cannot find numbers to support this sentence after s/MobilenetV2-Small/MobilnetV3-Small/"
Introducing MobileNetV3: Based on MNASNet, found by architecture search, we applied additional methods to go even further (quantization friendly SqueezeExcite & Swish + NetAdapt + Compact layers). Result: 2x faster and more accurate than MobileNetV2. Link: https://arxiv.org/abs/1905.02244 pic.twitter.com/jEFBeA67sR
-
-
-
Thank you for spotting the typo. The result that supports this sentence can be found in Table 3.
- Još 1 odgovor
Novi razgovor -
-
-
@midasIIITD @the_dhumketu@astitwa here they go again. -
Wow, many researchers are now working in this area.
Kraj razgovora
Novi razgovor -
-
-
Looking forward to using it as a Keras Application
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
Thanks! I'm curious about a detail. The paper suggests the last block of 5x5 convs has strides 1-2-1, but this deviates from MNASNet pattern more significantly than the 'two types of proposals'. Intended? I've tried to replicate with my generic mobilenet: https://github.com/rwightman/pytorch-image-models/blob/master/models/genmobilenet.py#L883 …
-
Good catch, Ross! It is a typing mistake, and the correct stride is indeed 2-1-1. We will update the arxiv paper with correct numbers soon. Thank you!
- Još 2 druga odgovora
Novi razgovor -
-
-
The paper is really really good. Thank you for this. Though there is one thing that isn't clear to me yet. How are the `bneck` layers arranged? Can you detail out the parts of bneck layer? It is hard to reproduce without exact details
-
bneck is the building block described in figure 4.
Kraj razgovora
Novi razgovor -
-
-
Please please consider releasing a pytorch version of the code with the pretrained weights ?
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.