11: Chapter 11 First Generation of African Beauty Techniques

"First, we need to build a facial database containing 100,000 images of people with diverse skin tones, with Black samples accounting for 40%. Then, we will use this data to retrain the face detection model."

"Second, we need to modify the Image Signal Processing pipeline." The flowchart was displayed with red revision markers clearly visible; it was almost like an operating manual, giving the engineers in the audience a hint.

"We will perform skin tone recognition at the RAW level, and apply different noise reduction parameters and detail enhancement strategies based on different skin tones. Black skin tends to lose yellow and red information in low light, so we need to protect that at an early stage."

"Third, and the most critical step..." At this point, Shen Fei paused slightly, knowing they would be deeply surprised by the following news.

"We also need to develop a deep learning-based skin tone restoration network."

The conference room went completely silent. In 2008, deep learning was not unheard of, but it was still a cutting-edge concept in academia. Although various models were hyped up, there were almost no real-world commercial applications... Even if there were, they were basically used to scam investors, and the actual models were at the level of a moron.

Transsion was a bankrupt small business setting such high goals. The good news: the boss knew a bit about technology. The bad news: not enough, it seemed; this kind of work would be even more tiring. Some engineers who had joined because of the salary now felt like backing out.

Seeing everyone's hesitation, Shen Fei laughed instead: "Rest assured, this deep learning model has already been developed, and the current results are quite good. Let me introduce it to you in detail."

Following this, amidst everyone's shock, Shen Fei opened another PPT and began a more detailed explanation.

"This network has 12 layers in total; the input is the preliminarily processed image, and the output is the result with optimized skin tone. We added a large number of multi-skin tone portraits to the dataset for fine-tuning. The network can automatically recognize race, lighting conditions, and shooting environments to personalize the optimization for each photo."

To further stimulate these treasures he had poached, Shen Fei brought out the real deal and displayed a portion of the code! The structure was exquisite!

"The algorithm complexity is controlled within an acceptable range. On a 300MHz ARM11 processor, it only takes 0.8 seconds to process an 8-megapixel photo. If reduced to 300,000 pixels, it only takes 0.1 seconds, which can fully meet the needs of real-time preview."

At this moment, the engineers in the audience were completely stunned. This Transsion seemed different from what was advertised... It really had technical strength. If this processing speed was called 'acceptable,' then this was god-tier. Even at Nokia or Motorola, it would take years to develop such a perfect algorithm. But Transsion had already brought it out.

An engineer raised his hand: "Boss Shen, the patent for this algorithm..."

"Global patents are already being applied for..." Shen Fei smiled. "Patents in China, the United States, Europe, and major African countries have already been submitted. Transsion will have exclusive usage rights for at least the next 20 years."

Chen Hai stood up excitedly: "Boss Shen, based on this algorithm alone, Transsion's photo quality in the African market will be able to crush all competitors. This is simply... This is simply a..."

"Killer app." Shen Fei finished for the excited Chen Hai.

Then another engineer suddenly realized and asked tentatively: "Boss Shen, this project doesn't actually need us to develop it as you said, but has it already been completed? That photo just now shouldn't be Photoshop, but rather... So what we are going to do next is just simple adaptation work."

It's easy talking to smart people. Shen Fei stood up and clapped: "That's right, this project has already been implemented. That photo just now was taken by a heavily modified Transsion T100. But having the algorithm is just the starting point; adaptation is equally important. I hope everyone can work with all your might so that we can capture the entire African market as soon as possible."

"Don't worry, Boss, we guarantee to complete the mission. It's just that we don't know if we can meet these big shots; we have some questions we'd like to ask."

Some engineers who originally wanted to coast were now infected by Transsion's hidden R&D capability... They also wanted to learn and take their skills to the next level.

At this moment, Shen Fei turned off the projector and turned on the lights in the conference room. "Everyone, no matter how strong the algorithm is, it needs to be carried on hardware. Our hardware..." Shen Fei looked at Chen Hai, "Engineer Chen, how is the progress on the engineering prototypes?"

Chen Hai was full of energy and immediately reported his work results: "The T80 engineering prototype is complete, with dual SIM dual standby, using the MT6223D. The chip quote is 3.2 USD, and all costs combined can be controlled within 20 USD. The T1 prototype is still being debugged; that 4-SIM 4-standby requires some optimization of the motherboard, and the battery compartment needs to be enlarged."

The MT6223D chip only supports calls and communication, with all other functions stripped away. It is positioned as an ultra-low-end product within MediaTek, but this chip is perfectly sufficient for the African market.

"Time!" "The T80 can enter small-batch trial production this weekend, with an initial batch of 5,000 units... The T1... will need at least another two weeks."

Shen Fei looked at the calendar: "Too slow. I'm adding 3 million to the R&D center's budget. If you don't have enough personnel, hire them; if you don't have enough equipment, buy it. In 10 days, I want to see a trial production batch of 100 units for each of the two prototypes. If there are no issues, we will start mass production."

"Boss, the money... we're running out of money." The new CFO Wang Shanghua and Procurement Manager Liu Fang said in unison.

In the past two days, for the progress of the T80 and T1 projects, they had purchased a lot of materials. To compress costs, everything was purchased in large quantities. They directly purchased 1 million MediaTek MT6223D chips and managed to cut 0.2 USD off the price... But MediaTek was also worried that Transsion wouldn't be able to pay, so they demanded a deposit before agreeing to ship... So, the current account is really...

"Don't worry about the money; the company will soon receive an additional capital injection."

"Then there's no problem!"

After the meeting, Chen Hai chased after Shen Fei, his eyes filled with a trace of admiration: "Boss Shen, that algorithm... was it really developed by your friend?"

Shen Fei looked at him: "Engineer Chen, don't dig too deep into some questions. You only need to know that I haven't been wasting my time over the past few years. The Transsion of now is not the Transsion of last year; we now have a technological moat. Next, we are going to turn it into a product barrier as quickly as possible."

He patted Chen Hai's shoulder: "In a month, the first batch of goods will be shipped to Nigeria. Whether we can win this battle depends on you all."

Over the next month or so, Transsion entered a period of frantic R&D work. The R&D center worked in 24-hour shifts, and the production line ran in three shifts. But all the employees had smiles on their faces; after all, Young Boss Shen was really paying! Overtime pay was sufficient, and even when working overtime at night, they would get paid before working... This didn't need any reminders from the production line team leaders; they would work desperately on their own and do their best to improve accuracy.