Blog

As men­tioned in my pre­vi­ous report, I have now bought the DJI Mini 3 Pro with the RC con­troller* in addi­tion to the DJI Mini 2 and the DJI Air 2S. I have already sold the Mini 2 in the meantime.

As a pho­tog­ra­ph­er, I am pri­mar­i­ly inter­est­ed in the qual­i­ty of the pic­tures I take with my drones. There are already many very pos­i­tive reviews on the Inter­net con­cern­ing the video capa­bil­i­ties of the DJI Mini 3 Pro. In the past, I used two dif­fer­ent drones for my pho­tos, depend­ing on the cir­cum­stances. One was the DJI Mini 2 as a small drone, which is more straight­for­ward to take and use because of its light weight and size. It can be flown legal­ly in many more regions than its heav­ier weight larg­er sis­ters. The price, how­ev­er, was low­er pho­to qual­i­ty due to the small­er 1/2.3″ sen­sor with its low­er res­o­lu­tion (12 megapix­els). How­ev­er, the avail­able panora­ma modes par­tial­ly com­pen­sat­ed for this.

After the first flight tests, I there­fore bought the larg­er DJI Air 2S as a sec­ond drone, which offers a res­o­lu­tion of 20 megapix­els on its 1″ sen­sor and thus rough­ly match­es the image qual­i­ty of a Sony RX-100 - which it actu­al­ly does. Unfor­tu­nate­ly, the use of this larg­er drone is sig­nif­i­cant­ly more restrict­ed by the EU drone regulations.

For more infor­ma­tion about the nec­es­sary require­ments and the reg­u­la­tions that apply in the EU, I would like to rec­om­mend my pre­vi­ous arti­cle “Pho­tog­ra­phy in 3 dimen­sions” . In order to use the DJI Air 2s, you will also need the so called “small drone license”, but you can get it online with­out any problems.

The DJI Mini 3 Pro

now seemed to be able to com­bine the advan­tages of both drones in one from its tech­ni­cal data. A first field report about the Mini 3 Pro with some exam­ples is already pub­lished on this website.

In order to fur­ther eval­u­ate the pho­to qual­i­ty of the DJI Mini 3 Pro, I have pre­pared a pho­to com­par­i­son series between the DJI Mini 3 Pro, the DJI Air 2S and, as a ref­er­ence, the Canon EOS R5. For­mal­ly, the DJI Mini 3 Pro even offers a high­er res­o­lu­tion than the EOS R5 (48 vs 45 megapixels).

DJI Mini 3 Pro vs DJI Air 2s vs Canon EOS R5

The sub­ject of the pho­to was the view from my win­dow. First, here are the results of the three pho­tos I took in quick suc­ces­sion. For com­pa­ra­bil­i­ty, I cropped the DJI Mini 3 Pro shot from 4:3 to the 3:2 for­mat of the oth­er cam­eras. I left the ISO at 100 for all pho­tos and oth­er­wise let the cam­era auto expose. As always, I shot all images in RAW for­mat (DNG or CR3) and then export­ed them using the default set­tings with Adobe Light­room Clas­sic at 2048 pix­els wide initially:

All in all, the qual­i­ty of the three cam­eras is not too bad, at least in the res­o­lu­tion of effec­tive­ly a good 3 megapix­els shown here. How does it look in the larg­er view?

Image details

For this pur­pose, I placed the images side by side in Adobe Light­room Clas­sic in high­ly mag­ni­fied com­par­i­son view. Below I show the cor­re­spond­ing screen­shots. First, the com­par­i­son of the DJI Mini 3 Pro with the DJI Air 2s. Because of the dif­fer­ent res­o­lu­tion (48 vs 20 megapix­els) of both images, I enlarged the image of the Mini 3 Pro to 200% and the image of the Air 2s to about 350% to achieve a com­pa­ra­bil­i­ty of the details. Here is a crop from the cen­ter of the above image:

On clos­er inspec­tion, the Mini 3 Pro shows a slight­ly high­er res­o­lu­tion in the details than the DJI Air 2s. This can be seen in the win­dow bars, for exam­ple. At the same time, how­ev­er, some arti­facts are notice­able in the area of the wall at the bot­tom of the image. I have the impres­sion that the DJI Mini 3 already applies sig­nif­i­cant de-nois­ing to the RAW internally.

Now the com­par­i­son with a full-frame sys­tem cam­era. The Canon EOS R5 offers a res­o­lu­tion of 45 megapix­els on a 24x36mm sen­sor. I used the RF 24-105 f/4L lens on the cam­era at 24mm focal length and open aper­ture. Of course, the com­par­i­son is very unfair. The lens alone costs more than the DJI Mini 3 Pro with RC controller:

There is no need to debate, the image of the EOS R5 is clear­ly more detailed, clear­er and sharper.

Here is a sec­ond detail from the above pho­to, which allows you to judge the sharp­ness a lit­tle bet­ter. I have cho­sen a sec­tion with the church top for this pur­pose in each case. With the fine­ly detailed weath­er­cock mount­ed on the church spire, the res­o­lu­tion of fine details can be judged even bet­ter. As before, I show on the left the image of the DJI Mini 3 Pro in 200% and on the right at first the image of the DJI Air 2s in about 350% magnification:

Here you can now clear­ly see that the DJI Mini 3 Pro offers a much high­er res­o­lu­tion of details than the DJI Air 2s. Of course, the Canon EOS R5 is even bet­ter here, as can be seen in the image below (both crops in 200% magnification):

Aber im direk­ten Ver­gle­ich schlägt sich die DJI Mini 3 Pro hier doch wieder ganz gut. Ich meine, dass hier das Ergeb­nis näher an der Canon EOS R5 als an der DJI Air 2s liegt.

Dynamic range

In addi­tion to the detail res­o­lu­tion of cam­era sen­sors, the usable dynam­ic range is also of inter­est. To check this, I light­ened the shad­ows in a dark area of the above image to the max­i­mum in Light­room by set­ting the “Shad­ows” slid­er to +100. The com­par­i­son to the DJI Air 2s and the Canon EOS R5, each at an adjust­ed mag­ni­fi­ca­tion, can again be seen below. As above, you can see the image sec­tion of the DJI Mini 3 Pro on the left:

In my opin­ion, the over­all result is real­ly respectable. The DJI Mini reveals more details in the shad­ow areas than the Air 2s, with­out a lot of noise there. The Canon EOS R5 is again, of course, much better.

And what else?

A clos­er look at the image details reveals anoth­er spe­cial char­ac­ter­is­tic of the Mini 3 Pro’s pho­tos, which I think can be explained by its spe­cif­ic sen­sor design. The cam­era of the Mini 3 Pro uses a so-called Quad Bay­er sensor.

To clar­i­fy the prob­lem, let’s take a brief excursion:

Bayer Sensor Design

Nor­mal­ly, the light-sen­si­tive pix­els of dig­i­tal cam­era sen­sors can­not dis­tin­guish col­ors; they only see grayscales. To dis­tin­guish col­ors, there­fore, col­or fil­ters are placed in front of the indi­vid­ual light-sen­si­tive pho­to­cells of the sen­sor chip, which then only allow light of the rel­e­vant col­or to pass through. Cor­re­spond­ing to the three col­or-sen­si­tive cones of our reti­na, these col­or fil­ters are red, green and blue (RGB). Most sen­sors use an 2x2 grid arrange­ment for this, which con­sists of one red, one blue, and two green fil­ters. Green is pro­vid­ed twice, since our eye is par­tic­u­lar­ly sen­si­tive in this area.

Wikipedia writes about this:

A Bay­er fil­ter mosa­ic is a col­or fil­ter array (CFA) for arrang­ing RGB col­or fil­ters on a square grid of pho­to­sen­sors. Its par­tic­u­lar arrange­ment of col­or fil­ters is used in most sin­gle-chip dig­i­tal image sen­sors used in dig­i­tal cam­eras, cam­corders, and scan­ners to cre­ate a col­or image. The fil­ter pat­tern is half green, one quar­ter red and one quar­ter blue. It is named after its inven­tor, Bryce Bay­er of East­man Kodak. Bay­er is also known for his recur­sive­ly defined matrix used in ordered dither­ing.

The col­or fil­ter arrange­ment then looks like this:

There­fore, it becomes clear that the sen­sor res­o­lu­tion of the indi­vid­ual col­ors must always be low­er than the total res­o­lu­tion. The cam­eras or the RAW con­vert­ers cal­cu­late the col­or of each indi­vid­ual pix­el with more or less suit­able algo­rithms from the col­or dis­tri­b­u­tion of the adja­cent pix­els, which nev­er fits to 100%. Effec­tive­ly, there­fore, the col­or chan­nel green offers only half and the oth­er col­or chan­nels only a quar­ter of the spec­i­fied sen­sor res­o­lu­tion. This is usu­al­ly not a big prob­lem, because our eye does not han­dle this in a fun­da­men­tal­ly dif­fer­ent way. Our reti­na con­tains sig­nif­i­cant­ly few­er col­or-sen­si­tive cells (approx. 6 mil­lion so-called cones) than bright­ness-sen­si­tive cells (approx. 110 mil­lion so-called rods).

Quad Bayer

How­ev­er, the DJI Mini 3 Pro push­es this prob­lem one step fur­ther. It has an even low­er res­o­lu­tion of the col­or chan­nels. Here is a com­par­i­son of the Bay­er and the Quad Bay­er arrange­ment of the col­or filters:

As you can see in the pic­ture, the indi­vid­ual col­or fil­ters have been sig­nif­i­cant­ly enlarged in the Quad Bay­er grid. They now cov­er 4 sen­sor fields each. This reduces the col­or res­o­lu­tion by a fac­tor of 4 com­pared to the clas­sic Bay­er matrix. And this is clear­ly vis­i­ble in the exam­ple images, for exam­ple in the bush­es. To illus­trate this effect, I am show­ing below a fur­ther enlarged sec­tion from the cen­ter of the image above with the bush­es. The left image of the DJI Mini 3 Pro has been enlarged in Light­room to 400%, the right one of the DJI Air 2s with the same detail to about 650%:

On the right image of the Air 2S, you can still see indi­vid­ual white dots in the bush­es that cor­re­spond to blos­soms. Despite the nom­i­nal­ly high­er res­o­lu­tion, these are only vis­i­ble as slight­ly brighter areas in the image of the Mini 3 Pro. The Mini 3 Pro’s ten­den­cy to pro­duce arti­facts is also inten­si­fied in this high mag­ni­fi­ca­tion. Over­all, in this case, the DJI Air 2s is a bet­ter choice despite its low­er resolution.

The dif­fer­ence is even more strik­ing when com­pared with the EOS R5 (both views mag­ni­fied 400%):

The EOS R5 is orders of mag­ni­tude bet­ter here, no ques­tion. Once again, the DJI Mini 3 Pro has obvi­ous­ly already pre-processed and de-noised the RAW files, which unfor­tu­nate­ly led to a sig­nif­i­cant loss of detail. How­ev­er, this should not be over­es­ti­mat­ed. After all, we are talk­ing about 400% enlarge­ments here. The result­ing total images would be a good 8 meters wide in a print cor­re­spond­ing to a usu­al 100 DPI screen resolution!

Why Quad Bayer?

The rea­son for this design mod­i­fi­ca­tion, which seems unrea­son­able at first glance, is that the sen­sor can alter­na­tive­ly read 4 indi­vid­ual pix­els at once with­out any prob­lems. Although the sen­sor res­o­lu­tion then drops to only 1/4 of the orig­i­nal num­ber of pix­els, each indi­vid­ual pix­el will be four times larg­er. This improves the sen­sor’s light sen­si­tiv­i­ty, noise per­for­mance and dynam­ic range at the expense of res­o­lu­tion. Thus, the DJI Mini 3 Pro then oper­ates at 12 megapix­els also in its pho­to mode. How­ev­er, since I will usu­al­ly use the drone in pho­tog­ra­phy in good weath­er and at its base ISO of 100, I would have pre­ferred a clas­sic Bay­er Matrix.

To the DJI Mini 3 Pro’s sal­va­tion, how­ev­er, it should be remem­bered that it is not yet direct­ly sup­port­ed by Light­room either. It is quite pos­si­ble that the Quad Bay­er data is not yet inter­pret­ed cor­rect­ly. Unfor­tu­nate­ly, the Mini 3 Pro is not yet sup­port­ed by DxO Pho­to­Lab or Pur­eRAW either. I hope that the AI-based Deep­PRIME algo­rithm avail­able there will improve the quality.

Résumé

I have mixed feel­ings about the DJI Mini 3 Pro’s pho­to results so far. Over­all, the detail res­o­lu­tion of the DJI Mini 3 Pro is bet­ter than that of the DJI Air 2s for suit­able sub­jects with­out large col­or vari­a­tions (as seen in the church tow­er exam­ple above).

The pho­tos are quite decent when viewed as a whole. How­ev­er, small col­ored details like the blos­soms in the bush­es in the exam­ple are some­times lost, and here the Air 2s is even part­ly supe­ri­or. In addi­tion, the RAW files of the Mini 3 Pro show clear arti­facts, which indi­cates inten­sive pre­pro­cess­ing and noise reduc­tion of the DNG files even at ISO 100. In any case, I would much pre­fer hon­est, unprocessed, noisy RAWs for photography.

These could then be processed with com­plex algo­rithms, such as the AI-based Deep­PRIME process in DxO Pur­eRAW or DxO Pho­to­lab, at the user’s choice and with pre­sum­ably much bet­ter results - pro­vid­ed that DxO also sup­ports the DJI Mini 3 Pro in the near future. How­ev­er, I strong­ly assume that this will be the case due to the wide accep­tance of the new drone. DxO added sup­port for the pre­de­ces­sor, the DJI Mini 2, a few months after the drone was released.

By no means does the drone even come close to the qual­i­ty of a full-frame sys­tem cam­era like the Canon EOS R5. How­ev­er, this was not to be expected.

I don’t know yet if it will replace my DJI Air 2s. In the video area, which I only use very rarely so far, the Air 2s is still clear­ly supe­ri­or to the Mini 3 Pro. But it also offers some advan­tages in pho­tog­ra­phy, such as long expo­sures with up to 10 sec­onds expo­sure time and pro­vides an expo­sure mode with aper­ture pri­or­i­ty at a fixed ISO val­ue, which I still miss a lot in the Mini 3 Pro.

The DJI Mini 3 Pro only uses the low­er 12 megapix­el res­o­lu­tion in the auto­mat­ic panora­ma modes I used fre­quent­ly before. The DJI Air 2s, on the oth­er hand, does this much bet­ter with its full res­o­lu­tion of 20 megapix­els. How­ev­er, with the Mini 3 Pro, this can at least be achieved man­u­al­ly con­trolled at 48 megapix­els. This is very cum­ber­some and I real­ly hope that the panora­ma ver­sions with 48 megapix­els will be added in a future firmware ver­sion. In addi­tion, the DJI Mini 3 Pro is basi­cal­ly the first mod­el to allow a full 360° spher­i­cal panora­ma, as it can tilt the cam­era up to 60°. The DJI Mini 2 and the Air 2s spher­i­cal panora­mas do not include a part of the image in the zenith because the gim­bal can only be tilt­ed upwards to a lim­it­ed extent.

*=Affil­i­at Link